diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml index fc55f1e3..398f0122 100644 --- a/.github/workflows/ci.yml +++ b/.github/workflows/ci.yml @@ -10,16 +10,32 @@ name: ci # Before this layout (run 44460876935, 2026-11-19): 9 minutes wall, 26 jobs, 87.5 runner-minutes; # gate shard 1 took 8 minutes (all 474 test files plus a recipe slice) and shots shards up to # 8.3 minutes (four `t3d` processes per folder, each re-importing and recompiling the recipe). -# Now: 16 shards (T3D_SHARD=k/16, CI_SHARDS in scripts/lib/shard.ts). Each runs its `bun test --shard` -# slice (balanced by scripts/lib/test-timings.json) and its units of recipe work, split by -# measured cost (scripts/lib/recipe-costs.json; greedy longest-first, scripts/lib/shard.ts): one -# folder's gate and its renders are separate units, renders take one process per folder +# Now: 18 shards (T3D_SHARD=k/18, CI_SHARDS in scripts/lib/shard.ts; with `site` that is 19 jobs, +# under the 20 a free account runs at once). Each runs its `bun test --shard` slice (balanced by +# scripts/lib/test-timings.json) and its units of recipe work, split by measured cost +# (scripts/lib/recipe-costs.json; greedy longest-first, scripts/lib/shard.ts): one folder's gate +# and its renders are separate units, renders take one process per folder # (scripts/shots-folder.ts); shard 1 also runs lint, types, build and pack. -# Measured 2026-11-19 (run 44474402275): 6.6 minutes push → green, shards 2.6-6.4 minutes, -# 76 runner-minutes. The floor is one unit: human_head's renders, because each of shot / hero / -# blueprint / icon compiles its LOD0 again (about 55 s each). Under 5 minutes needs the four -# renders to share one compile (a src/cli change; docs/maintenance.md "CI time"). When shards -# drift: `bun run scripts/recipe-costs.ts --tests`, then `--plan` to pick the shard count. +# Measured 2026-11-19 with 16 shards (run 44474402275): 6.6 minutes push → green, shards +# 2.6-6.4 minutes, 76 runner-minutes; the floor was human_head's renders, where each of shot / +# hero / blueprint / icon compiled LOD0 again. Since then the four renders share one compile +# (src/cli/compile-cache.ts). +# Measured 2026-11-19 with 18 shards (run 44492321426): 5.5 minutes push → green, 18 gate jobs of +# 155-299 s (mean 222 s), about 67 runner-minutes. The work was already small enough; the split +# was off by up to 35%, because the costs were dev-machine seconds with one guessed factor for +# renders (real factors: 0.9 to 2.3 per folder). Slowest shard 299 s: shard 8, whose rts_barracks +# renders took 153 s against 94 s in the costs. +# The costs are now that run's own GitHub seconds (`bun run scripts/recipe-costs.ts --from-run +# 44492321426`; scripts/recipe-costs-run.ts says what is measured and what is derived), with +# 16 s of overhead per job (checkout, setup, the stale check) and 26 s of shard 1's own steps. +# Measured after rebalancing (run 44494826233, 2026-11-19): 4.6 minutes push → green, slowest job 269 s. +# The plan it ran on: still 18 shards (the fewest whose worst job fits 240 s), every job +# about 222 s (`--plan`: 221-223 s; up to 237 s where the heaviest recipe shard also draws the +# 30 s test file src/kinds/creature-family.test.ts), about 4.3 minutes push → green (the slowest +# job plus about 30 s of queueing and the `verify` job), the same 67 runner-minutes. No single +# unit is the floor: the largest are face_demo's renders (171 s) and gate (166 s). Replace this +# paragraph with the next measured run. When shards drift: `--from-run ` on the commit a +# green run built, then `--plan` to pick the shard count (★). # Docs-only changes (scripts/ci-changes.ts: Markdown / docs/ outside recipes/) skip the recipe # gate and the renders; lint, types and every test still run, and `verify` still reports. @@ -43,9 +59,9 @@ jobs: strategy: fail-fast: false matrix: - shard: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16] + shard: [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18] env: - T3D_SHARD: ${{ matrix.shard }}/16 + T3D_SHARD: ${{ matrix.shard }}/18 steps: # Full history without old file contents (blob:none): enough for `git diff --name-only`. - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1 diff --git a/CHANGELOG.md b/CHANGELOG.md index ffdbe909..56b1bd2c 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -18,9 +18,15 @@ Several agents can share one machine through a resource queue, CI is balanced by - `build`, `shot`, `hero`, `review`, `critique`, `blueprint`, `icon`, `gaps`, `tune`, `finish`, `concept` and `neural run` take one CPU token; `hero --quality path` takes 8 and traces on the threads granted (same bytes). `blender review` and `blender session start|run|finish` take `blender`; Blender adapter jobs and jobs sent to a loopback model URL take `blender` / `gpu` while they run. A waiting command prints one `t3d queue: waiting …` line on stderr. - `bun run verify` holds one place in the queue for the whole gate; tests run with the queue off. - CI: `gate` and `shots` are one 16-shard `gate` job (26 jobs → 18) balanced by measured cost (`scripts/lib/recipe-costs.json`, `scripts/lib/test-timings.json`); `bun test` is split across the shards; docs-only changes skip the recipe gate and the renders. `verify` stays the single required check. -- `bun run shots` renders each folder's sheet, hero, blueprint and icon in one process, byte-identical to the four separate commands. +- `bun run shots` renders each folder's sheet, hero, blueprint and icon in one process and from one compile per asset and LOD (`human_head` renders 242 s → 88 s locally), byte-identical to the four separate commands. The shared compile is off everywhere else, so `t3d mcp` never serves an old compile. +- `bear_hybrid`, `wolf_hybrid` and `dragon_hybrid` are model-first with no CPU build: `t3d build` with a model output missing stops with one `T3D_NEURAL_MISSING` instead of building a code creature that fails QA. `modelChain` takes `required: true` and `ctx.neural` asks take `{ required: true }`; `ranger` and `human_head` keep their CPU fallback. +- CI: 18 gate shards with costs calibrated from a real run: `bun run scripts/recipe-costs.ts --from-run ` reads a green `ci` run's logs and rewrites the cost files in GitHub seconds; `--plan` marks the shard count to pick. Measured 5.5 min and 67 runner-minutes before rebalancing (jobs 155–299 s); measured 4.6 min push → green after (slowest job 269 s), under the 5-minute target for the first time since it was set. ### Fixed +- The npm package is 4.2 MB instead of 40.6 MB: model recipes' generated outputs (`recipes/*/neural/`) stay in git. 0.13.0 shipped them without the concept images they need, so they never built from the package. The `pack` gate step now fails when the tarball goes over 5.3 MB packed or 16 MB unpacked, holds a file over 1 MB, or holds anything under `recipes/*/neural/`. +- `T3D_NEURAL_MISSING` for a recipe inside the installed package says the package ships no model outputs and to clone the repo or copy the recipe folder; a `--require-neural` miss no longer says "using the CPU path". +- Model creatures: a rigger's terminal stub bone on a leg (UniRig, 6 mm past the toe) no longer shifts hip, stifle, hock and paw one joint down and gets the rig refused (`T3D_MODEL_SKELETON`). +- Critiques: every committed asset has a well-formed critique bound to its current renders, re-judged from the images, with a fix note under every score below 10. Malformed critiques had been hiding their scores from the gate. - CI's "committed renders match a fresh render" check never failed: its pathspec `recipes/*/out` matches no file under `out/`. It now uses `:(glob)recipes/*/out/**`, and a test proves the pathspec lists the committed outputs (D286). - `rts_faction`: committed renders refreshed; they still showed the tank's old belt weathering, which the broken check let through. diff --git a/CLAUDE.md b/CLAUDE.md index 68e5e233..caf1e4b3 100644 --- a/CLAUDE.md +++ b/CLAUDE.md @@ -91,7 +91,7 @@ Human names a thing → research specs + real photos into `recipes//README ## Coordination and CI - **Finish the job:** fix, test, document, run the gate. Decide and state decisions rather than asking what code/docs answer. Ask only before irreversible/outward-facing acts. - **Hive** (big tasks: `/feature`): one checkout, never git worktrees; only the top-level agent spawns; exclusive file sets, no git for agents; max 4 parallel; coordinator runs the full gate once at the end. Heavy commands take turns through the machine queue (ADR 017): a `t3d queue: waiting …` line is not a failure, `T3D_QUEUE_BUSY` / `T3D_QUEUE_TIMEOUT` mean a busy machine, not wrong code; never write `sleep`/`pgrep` wait loops, use `t3d queue status`. -- **CI:** free GitHub runners; target < 5 min (As of 2026-11: 9 min before cost-balanced shards; measured 6.6 min and 76 runner-minutes after, floor = `human_head`'s four renders each recompiling, see `docs/maintenance.md`). `ci.yml`: `gate` (16 shards, `T3D_SHARD=k/16`: each runs `bun run verify` — its `bun test --shard` slice plus its recipe units balanced by `scripts/lib/recipe-costs.json`; shard 1 also lint, types, build, pack — then `bun run shots` on its render units, failing on any change under `recipes/*/out` and uploading contact sheets when it fails; docs-only changes skip recipes and shots), `site`, and the aggregate `verify`, the one required check. `pages.yml`: gallery to developerz-ai.github.io/ts23d. `release-tag.yml` (idempotent) + `release.yml` (runs the full gate; refuses a tag not on a green `main` commit): release PR merge → tag → npm via OIDC (no tokens). Every action SHA-pinned; Dependabot moves pins. +- **CI:** free GitHub runners; target < 5 min (As of 2026-11: measured 5.5 min and 67 runner-minutes on 18 shards, slowest job 299 s; since then the costs are that run's GitHub seconds (`bun run scripts/recipe-costs.ts --from-run `), then measured 4.6 min with the slowest job at 269 s, see `docs/maintenance.md`). `ci.yml`: `gate` (18 shards, `T3D_SHARD=k/18`: each runs `bun run verify` — its `bun test --shard` slice plus its recipe units balanced by `scripts/lib/recipe-costs.json`; shard 1 also lint, types, build, pack — then `bun run shots` on its render units, failing on any change under `recipes/*/out` and uploading contact sheets when it fails; docs-only changes skip recipes and shots), `site`, and the aggregate `verify`, the one required check. `pages.yml`: gallery to developerz-ai.github.io/ts23d. `release-tag.yml` (idempotent) + `release.yml` (runs the full gate; refuses a tag not on a green `main` commit): release PR merge → tag → npm via OIDC (no tokens). Every action SHA-pinned; Dependabot moves pins. - **AI-maintained, for years untouched:** `.dz/maintainer/maintainer.yml` sets the developerz.ai lanes (review + merge on green for every author, triage, daily `health` and `security-release`, weekly `release-due` which cuts releases, monthly `improve` and `cleanup` which also moves the Bun pin and runs a January yearly review). [`docs/maintenance.md`](docs/maintenance.md) maps what decays over a decade to the lane that catches it; keep it true when a lane changes. Only owner settings (rulesets, npm trusted publisher, 1.0.0) stay human. ## Using the library in another project diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md index 2e2db0a9..bccec59f 100644 --- a/CONTRIBUTING.md +++ b/CONTRIBUTING.md @@ -33,7 +33,7 @@ bun run verify --json # for agents: { ok, failed, results, output | `test` | The `test` script in `package.json` (`src`, `scripts`, `examples`, `apps/site/src`; one 120 s per-test timeout) | `bun run test` | | `recipes` | Every recipe compiles at 3 LODs, passes QA, builds byte-identically twice, has `hero.png` + `sheet.png` + `manifest.json`; also runs each recipe's own `*.test.ts` | `bun run recipes` | | `build` | `dist/` JS + `.d.ts` | `bun run build` | -| `pack` | Packs the tarball, installs it in an empty project, imports every `exports` subpath under Bun and Node, runs `t3d build` | `bun run scripts/pack-smoke.ts` | +| `pack` | Packs the tarball, installs it in an empty project, imports every `exports` subpath under Bun and Node, runs `t3d build`, holds the tarball to the size budget in `scripts/pack-budget.ts` | `bun run scripts/pack-smoke.ts` | 11 steps, in this order (`STEPS` in `scripts/verify.ts`). An unknown `--only` name exits 2. In CI every shard runs its slice of `test` and `recipes`; the other steps run once, on shard 1. diff --git a/PUBLISHING.md b/PUBLISHING.md index 5bc8f945..e593c4b9 100644 --- a/PUBLISHING.md +++ b/PUBLISHING.md @@ -29,7 +29,7 @@ bun run release x.y.z ─► PR "chore: release x.y.z" ─► merge to main ─ | 1. Stamp | You / an agent | `bun run release x.y.z` sets `package.json` `version` (and `src/export/version.ts`), turns `## [Unreleased]` in `CHANGELOG.md` into `## [x.y.z] - ` (a fresh empty `[Unreleased]` stays on top), moves the link index at the bottom (`[Unreleased]` compares from `vx.y.z`, `[x.y.z]` compares from the previous tag) and regenerates `docs/changelog/` (a page per version plus the `README.md` index). `RELEASE_DATE=YYYY-MM-DD` overrides the date. Refused (`T3D_RELEASE`) when `x.y.z` is not above the current version, when `[Unreleased]` has no entries or no summary sentence, when `[x.y.z]` already exists, or when the `[Unreleased]` link does not compare from the latest version (`bun run changelog` rebuilds the index). Links and pages are generated, never hand-edited; the `conventions` gate step fails on drift | | 2. PR | You / an agent | `git switch -c release/x.y.z && git add -A CHANGELOG.md docs/changelog && git commit -am 'chore: release x.y.z' && gh pr create --fill`. Title **exactly** `chore: release x.y.z` | | 3. Merge | Reviewer | Merge to `main` with the title as the commit message (squash keeps it) | -| 4. `ci` | GitHub Actions | On the merge commit: `gate` (16 shards: `bun run verify`, then `bun run shots`, which fails if committed `recipes/*/out` are stale), `site` (`bun run site:build`). The `verify` job passes only when both succeeded; it is the one required check | +| 4. `ci` | GitHub Actions | On the merge commit: `gate` (18 shards: `bun run verify`, then `bun run shots`, which fails if committed `recipes/*/out` are stale), `site` (`bun run site:build`). The `verify` job passes only when both succeeded; it is the one required check | | 5. `release-tag.yml` | GitHub Actions | Runs when `ci` succeeds on a push to `main` whose head commit message starts with `chore: release `. Reads the version from `package.json`, checks the stamp, then, each step skipped when already done: annotated tag `vX.Y.Z` on that commit (a tag on another commit fails) → GitHub Release with that CHANGELOG section → `gh workflow run release.yml --ref vX.Y.Z -f version=X.Y.Z` unless npm already serves `X.Y.Z`. Re-running it finishes a half-done release | | 6. `release.yml` `check` | GitHub Actions | Refuses any ref but `refs/tags/vX.Y.Z`, a tag commit that is not on `origin/main`, and a commit with no successful `ci` push run (`gh run list --commit`); re-checks the stamp; runs the whole gate, `bun run verify`, unsharded (every step, recipes included) | | 7. `release.yml` `publish` | GitHub Actions, environment `npm-publish` | npm ≥ 11.5.1, `npm publish --access public --provenance` (its `prepack` script runs `bun run build`). Only this job holds `id-token: write`; it restores no cache | @@ -38,7 +38,9 @@ The dispatch in step 5 is needed because a tag pushed with `GITHUB_TOKEN` trigge ## What ships -`package.json` `files` and `exports` decide it; `bun run scripts/pack-smoke.ts` proves it. It packs the tarball (`prepack` builds `dist/`), installs it into an empty project and, as a consumer: builds a two-box recipe with `bunx -p @developerz.ai/ts23d t3d build` to QA PASS, imports every `exports` subpath under Bun and Node, type-checks them with `tsc` (`skipLibCheck: false`, no DOM lib), and fails if the tarball carries `*.test.ts`, `recipes/**/*.png` or `dist/**/*.map`, lacks `CLAUDE.md`, or misses an `exports` target. `exports` is an explicit list: the root, one subpath per module in `scripts/lib/tiers.ts` except `cli`, and `./package.json`. The library imports under Node ≥ 22 (the `default` condition, `dist/`); the `t3d` CLI and the examples need Bun. `AGENTS.md` is a symlink to `CLAUDE.md` and npm skips symlinks, so the package ships `CLAUDE.md` only. +`package.json` `files` and `exports` decide it; `bun run scripts/pack-smoke.ts` proves it. It packs the tarball (`prepack` builds `dist/`), installs it into an empty project and, as a consumer: builds a two-box recipe with `bunx -p @developerz.ai/ts23d t3d build` to QA PASS, imports every `exports` subpath under Bun and Node, type-checks them with `tsc` (`skipLibCheck: false`, no DOM lib), and fails if the tarball carries `*.test.ts`, `recipes/**/*.png`, `dist/**/*.map` or anything under `recipes/*/neural/`, lacks `CLAUDE.md`, misses an `exports` target, or breaks the size budget in `scripts/pack-budget.ts` (as of 2026-11: 5.3 MB packed, 16 MB unpacked, 1 MB per file; measured 4.2 MB / 12.7 MB, about 2,300 files). The failure names the largest files and the fix: exclude what a consumer does not need in `files`, or raise `PACK_BUDGET` with a reason in the diff. + +The model-built recipes' outputs (`recipes/*/neural/`: concept PNGs, image-to-3D GLBs, `record.json`; about 40 MB) stay in git, not in the tarball. From `node_modules`, `ranger` and `human_head` build their CPU tier (QA PASS) with a `T3D_NEURAL_MISSING` notice; `bear_hybrid`, `wolf_hybrid` and `dragon_hybrid` have no CPU build that passes QA, so `t3d build` stops with one `T3D_NEURAL_MISSING` error. In both cases the `fix:` says the package ships no model outputs: clone the repo to build the model tier, or copy the recipe folder and run `t3d neural run` on it. `exports` is an explicit list: the root, one subpath per module in `scripts/lib/tiers.ts` except `cli`, and `./package.json`. The library imports under Node ≥ 22 (the `default` condition, `dist/`); the `t3d` CLI and the examples need Bun. `AGENTS.md` is a symlink to `CLAUDE.md` and npm skips symlinks, so the package ships `CLAUDE.md` only. ## Versioning @@ -68,7 +70,7 @@ Repository → Settings → Environments → **New environment** `npm-publish`. ### 2b. Repository rules -- Ruleset on `main`: require a pull request and the `verify` status check (the `ci` aggregate job; one check, not the 25 shard jobs). +- Ruleset on `main`: require a pull request and the `verify` status check (the `ci` aggregate job; one check, not one per job: 18 `gate` shards and `site`). - Tag ruleset on `v*`: restrict creation to maintainers and the release workflow, so nobody else can mint a release tag. ### 3. Trusted publisher on npmjs.com diff --git a/README.md b/README.md index 0942a423..777dbb27 100644 --- a/README.md +++ b/README.md @@ -368,7 +368,7 @@ Runnable copy: [`examples/build-with-bun.ts`](examples/build-with-bun.ts) — ex ## Use it with any agent -**The docs ship inside the package, versioned with the code, readable offline.** After `bun add`, everything an agent needs is at `node_modules/@developerz.ai/ts23d/`: `CLAUDE.md` (the agent brain; the repo's `AGENTS.md` is a symlink to it and is not in the tarball), `llms.txt`, `docs/`, `recipes/` (worked examples), `CHANGELOG.md` and `.claude/{skills,agents,commands}`. +**The docs ship inside the package, versioned with the code, readable offline.** After `bun add`, everything an agent needs is at `node_modules/@developerz.ai/ts23d/`: `CLAUDE.md` (the agent brain; the repo's `AGENTS.md` is a symlink to it and is not in the tarball), `llms.txt`, `docs/`, `recipes/` (worked examples; the model recipes' `neural/` outputs stay in git: from the package `ranger` and `human_head` build their CPU tier, and `bear_hybrid`, `wolf_hybrid` and `dragon_hybrid`, which have no CPU build, stop with `T3D_NEURAL_MISSING` and a `fix:` that says to clone the repo), `CHANGELOG.md` and `.claude/{skills,agents,commands}`. Point your agent at them. Add to your project's `CLAUDE.md` or `AGENTS.md`: diff --git a/docs/architecture/defect-catalogue.md b/docs/architecture/defect-catalogue.md index 67d48db3..c8bcef14 100644 --- a/docs/architecture/defect-catalogue.md +++ b/docs/architecture/defect-catalogue.md @@ -260,7 +260,7 @@ again. A defect is not fixed until its check exists and fails on a reproduction. | D250 | t34: the turret slogan stood 0.8–1.1 cm off the armour; the LOD2 track showed as see-through dashes from the side | a flat decal 8 mm off a warped hex-turret plate; the LOD2 slab omitted its outboard and inboard edges | `connections` (`src/qa/connections-held.test.ts`) | shipped | | D251 | water_tower: pull-chain links hung 1 cm apart; ladder stays ended 1.9 cm short of the tank; both `clear('spout', …)` checks tested nothing | links spaced further apart than they are long; stays too short; the selector named an empty group | `connections` (`src/qa/connections-held.test.ts`) and `checkSelectors` (`src/qa/check-selectors.test.ts`) | shipped | | D252 | roadster_v10: 227 loose pieces: grille bars, splitter, plate lettering, door locks, hinges without pins, speaker grilles, lamp light guides, fan blades, seat adjuster, belts, dash stitch, switches, pedals | fittings placed a few mm to 1.6 cm off the bowed or leaning surfaces that carry them; hinge leaves stopped short of the swing axis | `connections` (`src/qa/connections-held.test.ts`) | shipped | -| D253 | `wolf_hybrid`'s skin webs into a stretched sheet between the tail and the right thigh at the hip's extremes (QA `deform` ×5.0–5.7 on LOD1/2, ×4.2 on the left) | the image-to-3D mesh fused the tail to the inner face of the right thigh; `unweb` cuts only the free legs and the tail's reach claims the thigh, so `isolate` weights it to the tail | declared `limitTolerances` (`hind_r_0` ×5.7, `hind_l_0` ×4.3) with the cause; ratchet in `recipes/wolf_hybrid/wolf_hybrid.test.ts` (at most 128 tail-to-leg edges); the fix is a per-bone `isolate` reach or a re-generated mesh (GPU) | open | +| D253 | `wolf_hybrid`'s skin webs into a stretched sheet between the tail and the right thigh at the hip's extremes (QA `deform` ×5.0–5.7 on LOD1/2, ×4.2 on the left) | the image-to-3D mesh fused the tail to the inner face of the right thigh; `unweb` cuts only the free legs and the tail's reach claims the thigh, so `isolate` weights it to the tail | declared `limitTolerances` (`hind_r_0` ×5.7, `hind_l_0` ×4.3) with the cause; ratchet in `recipes/wolf_hybrid/wolf_hybrid.test.ts` (at most 128 tail-to-leg edges); the fix is a per-bone `isolate` reach or a re-generated mesh (GPU) Tried 2026-11-19 and restored: a re-generated mesh with the tail swept back (z-image "wind" concept, Pixal3D) frees the tail from the hocks but not from the thighs and deforms worse (hips ×4.2–5.2, stifle volume 26–53 %); z-image draws a second tail for any raised-tail prompt; on the committed mesh, thigh capsules in `unweb`, `bake.cageFit` and a wider `isolate` feather leave ×4.1–4.6. Needs a per-bone reach or blend at the tail root in `isolate`, or a concept with one raised tail | open | | D254 | A known, tracked deform excess could only fail the build or be hidden by loosening the gate for everyone | `deform` had one global stretch limit (×4) and no declared exception | per-limit `stretch` + `why` (default ×4, cap ×6, `T3D_RECIPE_INVALID` otherwise), `limitTolerances` for limits a kind generates, listed in `manifest.json` `deformTolerances`; `src/qa/deform.test.ts` | shipped | | D255 | `t3d concept … --json` left stdout empty on a bad argument and its error had no `fix` | concept-cli printed its own JSON through console.log (sent to stderr under --json) with code and message only | `src/cli/run-json-errors.test.ts` (every command, bad argument, one ok:false envelope with error.code + error.fix) | shipped | | D256 | After a host handled one SIGINT, tools started later survived Ctrl-C | the reaper listened `once` but stayed marked installed | `src/neural/tool-reap.test.ts` (two SIGINTs with a host handler) | shipped | @@ -299,6 +299,11 @@ again. A defect is not fixed until its check exists and fails on a reproduction. | D289 | `t3d queue run -- bun --help` printed t3d's help and never started the command; a `--json` after `--` turned t3d's own output into an envelope | global flags were looked for in the whole line, past the `--` that hands the rest to the queued command | `--help` and `--json` are read only before `--`; `src/cli/queue/cmd.test.ts` | shipped | | D290 | The test said a ticket outlives its owner while a tool runs, but passed with child-pid liveness removed | it never ended the owner and its tool (`true`) had already exited | a real long-lived tool, the owner marked dead, a contender that stays blocked until the tool ends; `src/cli/queue/acquire.test.ts` | shipped | | D291 | With `T3D_QUEUE=wait` exported, ordinary CLI tests queued behind real jobs; `T3D_QUEUE=fail bun run verify --json` on a full queue printed no JSON | the test preload kept an exported mode; the gate runner let the acquisition error escape its reporting | the preload always sets `off`; the gate reports the error (with its fix) in both output modes; `scripts/test-setup.ts`, `scripts/verify.ts` | shipped | +| D292 | The 0.13.0 npm tarball was 40.6 MB packed / 52 MB unpacked, and its model recipes (bear_hybrid, ranger) failed to build from node_modules with `T3D_NEURAL_OUTPUT_INVALID` "output-0.png is missing" | `files` in package.json shipped each model recipe's `neural/` folder (about 39 MB of GLBs and records) while the PNG exclusion dropped the concept images those records need; no size check on the tarball | the `neural/` folders stay out of the package; a size budget in the pack step (5.3 MB packed, 16 MB unpacked, 1 MB per file); `scripts/pack-budget.test.ts` | shipped | +| D293 | A quadruped whose auto-rigger (UniRig) gave a hind leg five bones ending in a 6 mm stub was refused with `T3D_MODEL_SKELETON` (paw bone would turn 85°) | `rigTopology` took a limb chain's last four bones as hip → toe, so the stub became the toe and the hip landed on the stifle | a terminal bone shorter than 10 % of the mean of the chain's other bones is dropped while the chain has more than four (shortest real toe in the committed rigs: 14 %); `src/kinds/model-joints.test.ts` | shipped | +| D294 | From the installed npm package, a model recipe's `T3D_NEURAL_MISSING` told the user to run `t3d neural run` and commit the outputs, advice for a checkout the user does not have | the miss's fix text was one string in `neuralCache`; nothing knew the recipe sat in a package that ships no model outputs | `provideRecipeNeural` detects a recipe inside the installed package with no `neural/` folder and passes `missingFix` (clone the repo, or copy the folder and run `t3d neural run`); `src/cli/neural-store.test.ts`, `src/neural/recipe-cache.test.ts` | shipped | +| D295 | `t3d build` of `bear_hybrid`, `wolf_hybrid` or `dragon_hybrid` without their model outputs printed a notice, then built a different asset that failed QA (LOD0 100–146k triangles against 30k, proportions, seams) | `modelChain` returned null on any miss and `defineModelCreature` built the code creature of the spec, which these recipes never held to their budget or box | `modelChain({ required: true })`: an offline (build) cache throws one `T3D_NEURAL_MISSING` on its miss; the three recipes set it; `src/kinds/model-chain.test.ts`, `src/neural/recipe-cache.test.ts` | shipped | +| D296 | Sharing a compiled asset across commands in one process kept gigabytes alive (roadster_v10: 2.7 GiB after GC, 8.7 GB peak) | QA and render memos are keyed on the parts array and mesh objects, so they live as long as the compiled asset does | `compileOnce` hands each caller a view (new asset, parts and mesh objects over the same typed arrays); `src/cli/compile-cache.test.ts` | shipped | Related: [errors.md](errors.md) (every `T3D_*` code), [16-art-as-text](../idea/16-art-as-text.md) (measurable art definitions), `docs/idea/06-qa.md` (gate table). diff --git a/docs/architecture/errors.md b/docs/architecture/errors.md index c9169351..ec6d3f03 100644 --- a/docs/architecture/errors.md +++ b/docs/architecture/errors.md @@ -96,7 +96,7 @@ T3D_PART_UNKNOWN: part 'gun_barrel' parents to 'gun', which does not exist yet | `T3D_QA_SWIM_RECOIL` | `swim` QA | The lateral centre of mass swings over 0.5 % of body length | Remove recoil at the root when baking | | `T3D_QA_SWIM_BURST` | `swim` QA | `burst` changes tail amplitude or wavelength over 15 % from `cruise` | Raise frequency only | | `T3D_QA_SWIM_BEAT` | `swim` QA | `cruise` or `burst` travels but the tail does not beat | Drive the spine with a travelling wave; the root only follows | -| `T3D_NEURAL_MISSING` | neural recipe cache (planned, [plan 101](../plans/2026/11/15/101-asset-realism-programme/overview.md) slice 13) | A recipe asked a neural task whose output is not in `recipes//neural/` (ADR 014). A notice and CPU-tier fallback by default; an error only with `--require-neural`. Also when a request's input is a local output (`keep: 'local'`) that is not on this machine: the request is not sent | `t3d neural run ` with a configured adapter, then commit the cache; for an absent local input, re-run or copy the upstream output on this machine | +| `T3D_NEURAL_MISSING` | neural recipe cache (planned, [plan 101](../plans/2026/11/15/101-asset-realism-programme/overview.md) slice 13) | A recipe asked a neural task whose output is not in `recipes//neural/` (ADR 014). A notice and CPU-tier fallback by default; an error with `--require-neural`, and in any build of a model-first recipe whose chain is `required: true` (`modelChain`; `ctx.neural(…, { required: true })`: `bear_hybrid`, `wolf_hybrid`, `dragon_hybrid`), which has no CPU build to fall back to. Also when a request's input is a local output (`keep: 'local'`) that is not on this machine: the request is not sent | `t3d neural run ` with a configured adapter, then commit the cache; for a recipe inside the installed npm package (which ships no `recipes/*/neural/`): clone https://github.com/developerz-ai/ts23d and build there, or copy the recipe folder into your project and run `t3d neural run` on the copy; for an absent local input, re-run or copy the upstream output on this machine | | `T3D_NEURAL_FAILED` | neural adapters (planned, [plan 101](../plans/2026/11/15/101-asset-realism-programme/overview.md) slice 13) | The adapter or model server failed: network, HTTP status, timeout, job error. The message names the adapter, task and server message | Start or fix the server the adapter points at; check `t3d neural ls` | | `T3D_NEURAL_OUTPUT_INVALID` | `checkNeuralResult`; per-task output validation (planned, [plan 101](../plans/2026/11/15/101-asset-realism-programme/overview.md) slice 13/15) | A model returned no output, empty bytes, a mime the task does not allow, a score outside 0..10, or an output failing its task's validation (mesh, texture, rig, motion). The message names which check failed | Make the adapter return the task's file type (`NEURAL_MIMES`) with real bytes; fix the model output the message names; for a reply over the cap raise the adapter's `maxBytes` in `t3d.config.ts` (`http(url, model, { maxBytes: 256 * 1024 * 1024 })`) | | `T3D_NEURAL_INPUT` | `t3d neural run` input-image rules (`checkInputImage`, [plan 101](../plans/2026/11/15/101-asset-realism-programme/overview.md) slice 15) | A `mesh` request's image fails a rule before anything is sent to a model: no plain background, not exactly one subject, under 60 % of the frame, cropped, shorter side under 512 px, blurred, or not a PNG. Cluttered photos give broken AI meshes (D55) | Generate a clean concept: `t3d concept ` (plain background, ¾ view, full frame), or strip the background with an image/segment model; `--allow-bad-input` overrides and records it in record.json | diff --git a/docs/idea/08-neural-adapters.md b/docs/idea/08-neural-adapters.md index 3945c11c..168af468 100644 --- a/docs/idea/08-neural-adapters.md +++ b/docs/idea/08-neural-adapters.md @@ -95,7 +95,7 @@ recipes//neural//output-0.png | Lookup in a build | By request (`requestKey`), optionally pinned to a model; lowest key wins when several models answered | | Cap | 12 MB of committed outputs per recipe; `fill` and this repo's recipes gate refuse past it (`T3D_NEURAL_FAILED`). Local outputs do not count | | Hit | No GPU, no network. Files and record are re-hashed and re-validated on load | -| Miss | `{ fallback: true }` + a `T3D_NEURAL_MISSING` notice naming `t3d neural run`; the recipe takes its CPU path | +| Miss | `{ fallback: true }` + a `T3D_NEURAL_MISSING` notice naming `t3d neural run`; the recipe takes its CPU path. An ask with `required: true` (a model-first asset with no CPU build) is a `T3D_NEURAL_MISSING` error in a build instead; `t3d neural run` still lists it as a request to fill | | `--require-neural` | A miss is an error | | Fill | `t3d neural run `: compiles LOD 0-2, collects the misses, runs each through the task's configured adapter, writes the key directory | | Licence | Recorded in `record.json` and the credits, never gated for your builds. This repository's recipes gate refuses committed outputs whose licence does not allow publishing (`publishable`, ADR 015) | diff --git a/docs/idea/09-kinds.md b/docs/idea/09-kinds.md index 052f0659..3a392866 100644 --- a/docs/idea/09-kinds.md +++ b/docs/idea/09-kinds.md @@ -149,7 +149,7 @@ as one textured part on every LOD (ADR 016); the code creature is the CPU fallba | Call | File | Does | |---|---|---| -| `modelChain(ctx, { concept, surface, spec, steps: (rig) => […], … })` | `model-chain.ts` | Four cached asks, each named by the previous output's sha256: `image` → `mesh` (kept local: raw meshes are 70–100 MB, a build never sees them) → `blender` PROXY (clean → 6 000-triangle shell: the committed geometry TS fits on) → `blender` FINAL (`transform(fit)` first, then the recipe's steps built from the rig, with the fitted skeleton as input so Blender can `bind`). `null` on any miss. A plain `steps` array is one pass (rigged models) | +| `modelChain(ctx, { concept, surface, spec, steps: (rig) => […], … })` | `model-chain.ts` | Four cached asks, each named by the previous output's sha256: `image` → `mesh` (kept local: raw meshes are 70–100 MB, a build never sees them) → `blender` PROXY (clean → 6 000-triangle shell: the committed geometry TS fits on) → `blender` FINAL (`transform(fit)` first, then the recipe's steps built from the rig, with the fitted skeleton as input so Blender can `bind`). `null` on any miss; with `required: true` (a recipe whose code creature is not a shippable asset) a miss in a build is `T3D_NEURAL_MISSING` instead. A plain `steps` array is one pass (rigged models) | | `modelRig(mesh, spec, { symmetric, drop })` | `model-rig.ts` | Everything TS derives from one mesh, memoised: fit, registered mesh, fitted skeleton, clips, limb capsules for `unweb`, `at(bone)` for focus spheres. `symmetric` (exact side-profile concepts; `true`, or '+' / '-' for the half the camera saw): puts the plane midway between the traced left and right legs, takes the far legs off, mirrors the kept half and forces a left/right symmetric skeleton; `side` feeds the `symmetrize` step and `far` the `amputate` step before it | | `modelFit(mesh, bones)` | `model-fit.ts`, `model-flesh.ts` | Registers a mesh of any yaw and size onto the plan's flesh capsules: 12 yaw starts, a two-sided trimmed ICP solved upright, scale set by the sourced dorsal height, levelled (≤ 6°) and grounded. Returns the similarity, a row-major 4×4 and the residual | | `fitSkeleton(mesh, bones, prefix)` | `model-skeleton.ts`, `model-legs.ts`, `model-ends.ts`, `model-sections.ts` | Fits the plan's armature INTO the mesh (its pose is not our rest pose): legs traced from the ground up along their medial axis, spine under the dorsal line, the head wherever the model turned it, the tail to its tip. Same names, order and parents | diff --git a/docs/maintenance.md b/docs/maintenance.md index 7d1c2e80..c4cbec6c 100644 --- a/docs/maintenance.md +++ b/docs/maintenance.md @@ -38,7 +38,7 @@ gate; a red gate is fixed at the root cause. | Docs | claims drift from code; links and badges rot | `conventions` gate (error codes, defect rows); `cleanup` rows `links`, `docs-drift` | rewrite the claim to match the code | | Plans and roadmap | plans left `in_progress`, roadmap ticks that lie | `cleanup` row `plans`; `improve` row `claims` | resume, or mark `superseded` with the reason | | Repo and package weight | renders and PNGs pile up | `cleanup` row `weight` (compared month to month) | `files` exclusions, smaller PNGs; history rewrite only by a human | -| CI time | recipes and tests grow; the push → green target is under 5 minutes. Measured 2026-11-19 (run 44460876935, old layout): 9 minutes, 26 jobs, 87.5 runner-minutes. Now 16 `gate` shards, each running its `bun test --shard` slice and its recipe units split by measured cost; measured 2026-11-19 (run 44474402275): 6.6 minutes, 18 jobs, 76 runner-minutes, slowest shard 6.4 minutes. The floor is one unit: `human_head`'s renders, because `shot`, `hero`, `blueprint` and `icon` each compile LOD0 again (about 55 s each) | nothing automatic yet: the target lives in `ci.yml`'s header; `improve` should compare the median `ci` run time on `main` month to month, and refresh the costs when shards drift (`bun run scripts/recipe-costs.ts --tests`, then `--plan` to pick the shard count; the shots step logs seconds per folder) | done: cost-balanced shards (`scripts/lib/shard.ts`, `scripts/lib/recipe-costs.json`), tests split across shards, renders in one process per folder (`scripts/shots-folder.ts`), docs-only changes skip recipe work. Next: share one compile across a folder's four renders (a `src/cli` compile cache or one multi-output command; `human_head` renders would drop from about 240 s to about 90 s locally); never by dropping a check | +| CI time | recipes and tests grow; the push → green target is under 5 minutes. Measured 2026-11-19 (run 44460876935, old layout): 9 minutes, 26 jobs, 87.5 runner-minutes. With 16 cost-balanced `gate` shards (run 44474402275, 2026-11-19): 6.6 minutes, 18 jobs, 76 runner-minutes, slowest shard 6.4 minutes; the floor was `human_head`'s renders, where `shot`, `hero`, `blueprint` and `icon` each compiled LOD0 again (about 55 s each). Then a folder's four renders shared one compile and there were 18 shards (run 44492321426, 2026-11-19): 5.5 minutes, 18 gate jobs of 155 to 299 s (mean 222 s), about 67 runner-minutes; the work fitted, the split was off by up to 35% because the costs were dev-machine seconds with one guessed factor for renders (measured: 0.9 to 2.3 per folder). Now the costs are that run's own GitHub seconds, with 16 s of overhead per job. Measured after rebalancing (run 44494826233, 2026-11-19): 4.6 minutes push → green, slowest job 269 s. The plan it ran on: 18 shards, every job about 222 s (at worst 237 s), about 4.3 minutes push → green, the same 67 runner-minutes; no single unit is the floor (largest: `face_demo`'s renders 171 s and gate 166 s) | nothing automatic yet: the target lives in `ci.yml`'s header; `improve` should compare the median `ci` run time on `main` month to month, and when shards drift recalibrate from a green run on the commit it built (`bun run scripts/recipe-costs.ts --from-run `, then `--plan` to pick the shard count, marked ★) | done: cost-balanced shards (`scripts/lib/shard.ts`, `scripts/lib/recipe-costs.json`), tests split across shards, renders in one process per folder (`scripts/shots-folder.ts`) with one compile per asset and LOD (`src/cli/compile-cache.ts`), docs-only changes skip recipe work, costs calibrated from a CI run's logs (`scripts/recipe-costs-run.ts`). Next: record the next measured run here and in `ci.yml`'s header; print seconds per folder in the `recipes` step (today a folder's gate cost is the step's time split by the previous costs); preload each shard with its `bun test --shard` slice (2.8 to 30 s, split independently today); then the head recipes (`face_demo` renders 171 s and gate 166 s, `human_head` gate 149 s) and `rts_barracks`' nine path-traced heroes (153 s), which cap how few shards can meet the target; never by dropping a check | | Gallery | three.js / Vite majors break the viewer | Dependabot + the `site` CI job; `health` row `pages-deployed` | port the viewer; the site stays a gallery, never a game | | The policy itself | schema changes on developerz.ai | the platform validates `maintainer.yml` on ingest | update the file to the published schema | diff --git a/llms.txt b/llms.txt index feeb3889..f14e8c50 100644 --- a/llms.txt +++ b/llms.txt @@ -3,7 +3,7 @@ > @developerz.ai/ts23d (TypeScript to 3D) — a free, deterministic 3D asset generator driven by coding agents. Assets are TypeScript modules (never JSON) built with a DSL: `defineAsset({ name, kind, real, sources, budget, build(ctx) })`, where `build` adds named, pivoted, animatable parts from primitives (`box`, `lathe`, `cylinder`, `prism`, `beam`, `planar`) or from SVG drawings written as text (`extrudeSvg`, `latheSvg`, `visualHull` front/side silhouettes), places detail with `place`/`merge`, composes other assets with `ctx.use`, and gates detail on `ctx.lod`. It compiles to glTF/GLB for any engine (Unity, Unreal, Godot, three.js, Babylon) — ts23d generates assets only, the engine is yours — (3 LODs, one shared trim-sheet material, baked motion), fails a QA gate on floating parts, budget overruns and wrong real-world proportions, and renders a zero-GPU six-view contact-sheet PNG so a vision-capable agent can check its own work. The CLI needs Bun; the library imports under Node ≥ 22. Zero runtime dependencies, MIT. Status 0.x: anything unbuilt throws T3D_NOT_IMPLEMENTED with the roadmap slice that builds it. Optional tier: any model or tool through a `NeuralAdapter` (no GPU? `kubeezAdapter` for images, a partner); outputs are cached inputs, so builds stay deterministic without them. ## Offline (inside your project) -The npm package ships the docs, skills, recipe sources and examples, versioned with the code — read them without internet (As of 2026-11: recipe `out/` renders, PNGs, tests and source maps are not in the tarball; read those on GitHub): +The npm package ships the docs, skills, recipe sources and examples, versioned with the code — read them without internet (As of 2026-11: recipe `out/` renders, PNGs, `recipes/*/neural/` model outputs, tests and source maps are not in the tarball, which stays under 5.3 MB; read those on GitHub, and clone the repo to build a model recipe's model tier: from the package `ranger` and `human_head` build their CPU tier, while `bear_hybrid`, `wolf_hybrid` and `dragon_hybrid` stop with `T3D_NEURAL_MISSING`): - `node_modules/@developerz.ai/ts23d/CLAUDE.md` — rules and the loop (the repo's `AGENTS.md` is a symlink to it and is not shipped) - `node_modules/@developerz.ai/ts23d/docs/` — DSL, IR, materials, agent loop, QA, kinds, decisions, roadmap - `node_modules/@developerz.ai/ts23d/recipes//` — idiomatic recipes, each with its research dossier (`README.md`); committed hero + contact sheet live in the GitHub repo under `recipes//out/` @@ -69,7 +69,7 @@ The human names a thing ("Tiger I, mid-production 1943"). The agent researches s - Sound: `sound.impact|engine|mechanism|ui` (`@developerz.ai/ts23d/audio`) — deterministic WAVs from material, size, cylinders/RPM, part motion; positional `KHR_audio_emitter` (docs/idea/17-sound.md). -- Tiers (as of 2026-11): CPU tier builds byte-identical with no GPU; the optional model tier plugs any model in through `NeuralAdapter` (`t3d neural ls|run`, outputs committed under `recipes//neural//` with `record.json`; `T3D_NEURAL_MISSING` falls back to the CPU path). `t3d doctor` / `t3d setup tools|gpu-min|gpu-full` / `t3d server start|stop|status` install and check it (docs/idea/08-neural-adapters.md). Organic assets (creatures, humanoids; As of 2026-11) are model-first and Blender-first: concept image (exact side profile for a quadruped, A-pose front for a biped, flat light) -> image-to-3D mesh (local GPU Pixal3D / TRELLIS.2, or Meshy as a hosted, paid option; `keep: 'local'`) -> recipe-written `blender` steps (retopo shell, uv, bake, bind, repose, weights, lods; ops and gates in `examples/blender/README.md`) -> `importModel` -> one textured part per LOD (`TextureSet`, ADR 016; one material per set; `budget.textureSets`). Code keeps scale, skeleton, clips, budgets and QA (gate `model`: `T3D_MODEL_FIT|SKELETON|TEXTURE|DOUBLED`). Kinds: `modelChain`, `defineModelCreature`, `riggedChain`, `defineModelCharacter`. Look with `t3d blender review --out ` (needs Blender; never cached); fill the cache with `t3d neural run` (paid adapters print credits; `--yes-spend` above 50; committed cache <= 12 MB per recipe). Shipped assets: `recipes/ranger` (rigged humanoid, `defineModelCharacter`) `recipes/wolf_hybrid` (quadruped, `defineModelCreature`: Pixal3D mesh, UniRig joints, walk and trot) and `recipes/bear_hybrid` (plantigrade quadruped, walk); the dragon model tier is planned. Limits: fused fingers, sculpted-clump hair, no jaw or blink on a fused head. Skills: `creature-artist`, `character-artist`. +- Tiers (as of 2026-11): CPU tier builds byte-identical with no GPU; the optional model tier plugs any model in through `NeuralAdapter` (`t3d neural ls|run`, outputs committed under `recipes//neural//` with `record.json`; `T3D_NEURAL_MISSING` falls back to the CPU path, or is an error for a chain declared `required: true`). `t3d doctor` / `t3d setup tools|gpu-min|gpu-full` / `t3d server start|stop|status` install and check it (docs/idea/08-neural-adapters.md). Organic assets (creatures, humanoids; As of 2026-11) are model-first and Blender-first: concept image (exact side profile for a quadruped, A-pose front for a biped, flat light) -> image-to-3D mesh (local GPU Pixal3D / TRELLIS.2, or Meshy as a hosted, paid option; `keep: 'local'`) -> recipe-written `blender` steps (retopo shell, uv, bake, bind, repose, weights, lods; ops and gates in `examples/blender/README.md`) -> `importModel` -> one textured part per LOD (`TextureSet`, ADR 016; one material per set; `budget.textureSets`). Code keeps scale, skeleton, clips, budgets and QA (gate `model`: `T3D_MODEL_FIT|SKELETON|TEXTURE|DOUBLED`). Kinds: `modelChain`, `defineModelCreature`, `riggedChain`, `defineModelCharacter`. Look with `t3d blender review --out ` (needs Blender; never cached); fill the cache with `t3d neural run` (paid adapters print credits; `--yes-spend` above 50; committed cache <= 12 MB per recipe). Shipped assets: `recipes/ranger` (rigged humanoid, `defineModelCharacter`) `recipes/wolf_hybrid` (quadruped, `defineModelCreature`: Pixal3D mesh, UniRig joints, walk and trot) and `recipes/bear_hybrid` (plantigrade quadruped, walk); the dragon model tier is planned. Limits: fused fingers, sculpted-clump hair, no jaw or blink on a fused head. Skills: `creature-artist`, `character-artist`. - Blender (as of 2026-11): `t3d blender session` keeps one Blender open for chained step batches with undo; the `stroke` step brushes an existing mesh headless; assets still come from replaying the finished program with `t3d neural run`. The GPL scripts are not in the npm package: pass `--scripts /examples/blender`. - Bodies beyond four legs (as of 2026-11): `defineBodyPlan` (fish, octopus, hydra, snake), baked by `swimClips`, `octopusClips`, `hydraClips`, `snakeClips`; `medium: 'water'` enables the `swim` QA gate. - Realism tools: `photoMatch` + `solveCamera` + `silhouetteIoU`; `t3d critique` calibrated score (flagships near 6 of 10, target 8.5, floor not yet enforced); sculpt engine (cubic kernels, strokes, `fair()`, VDM stamps); `creature()` with bones, fur cards, `skinWeights`, glTF skins and clips (walk/trot/gallop), `skin`/`deform`/`gait` QA; `t3d hero --quality path` (CPU path tracer); `t3d finish --hero` (normal/AO/height/ORM bake); Blender, gltfpack, Basis KTX2 and Godot tool adapters. diff --git a/package.json b/package.json index 335f0d2d..1e25b1af 100644 --- a/package.json +++ b/package.json @@ -130,6 +130,7 @@ "!recipes/**/*.test.ts", "!recipes/**/*.png", "!recipes/*/out", + "!recipes/*/neural", "!docs/media", "examples", "!examples/neural-server", diff --git a/recipes/amur_tiger/out/amur_tiger/critique.json b/recipes/amur_tiger/out/amur_tiger/critique.json index de473171..60066324 100644 --- a/recipes/amur_tiger/out/amur_tiger/critique.json +++ b/recipes/amur_tiger/out/amur_tiger/critique.json @@ -1,141 +1,71 @@ { "asset": "amur_tiger", "reviewer": "agent", - "overall": 6, + "overall": 6.1, "scores": { "silhouette": { "score": 7, - "reason": "Reads as a big cat / tiger from the side and 3/4: level back, low head, deep chest, digitigrade legs, J-hanging tail with a black tip (silhouette IoU LOD2 0.93-0.95). Against the Schönbrunn side photos the trunk is a little long and tubular, the neck long, the hind thigh shallow, and the ears read small and slightly 'mouse-cup' from the front; the winter ruff barely changes the outline.", + "reason": "Reads as a big cat / tiger from the side and 3/4: level back, low head, deep chest, digitigrade legs, J-hanging tail with a black tip. The trunk is still long and tubular beside the Schönbrunn photos, the hind thigh shallow, and the ears read as thin cups: in hero.png the near ear is a pale disc and the far ear a black spike, so neither reads as a rounded tiger ear.", "fixes": [ - "shorten the trunk masses ~5% and deepen the chest at the brisket so the waist tuck reads", - "grow the cheek ruff lobes outward/down (ruff.ts) so the front silhouette widens below the ears like an Amur in winter", - "raise and enlarge the hamstring/quadriceps masses so the haunch reads as a powerful thigh in side view" + "shorten the trunk masses ~5% and deepen the brisket in torso.ts so the waist tuck reads in side view", + "thicken the ear shell and round its rim in ears.ts so it keeps a rounded outline edge-on (no disc / spike at 3/4)", + "enlarge the hamstring/quadriceps masses in legs.ts so the haunch reads as a powerful thigh" ] }, "proportions": { "score": 7, - "reason": "Sourced numbers hold in code: head+body 1.957 m (1.95), tail 0.987 m (0.99), withers 1.013 m (0.8-1.1), mass 237 kg (Mazák 180-306), limb allometry 264 kg. By eye beside the photos the head is still slightly small and low, the muzzle a touch long and the forearms a little thin for an Amur male.", + "reason": "Sourced numbers hold in code (head+body ~1.96 m, tail ~0.99 m, withers ~1.0 m on the 2.2 × 0.5 × 1.0 m sheet). By eye beside photos the head is slightly small and low, the muzzle a touch long and the forearms thin for an Amur male.", "fixes": [ - "thicken the forearm ellipsoids ~10% (legs.ts)", - "shorten the muzzle 1 cm and widen the whisker pads (face.ts)" + "thicken the forearm masses ~10% in legs.ts", + "shorten the muzzle 1 cm and widen the whisker pads in face.ts" ] }, "materials": { "score": 5, - "reason": "Everything is vertex colour on the plain strip: the coat reads as painted velvet, not fur; stripes are crisp but their edges stair-step at the 1.2 cm body cell; nose leather, eyes and claws separate by hue only. No fur cards/shells (planned M4), no whiskers.", + "reason": "Vertex colour on the plain strip: the coat reads as painted velvet, not fur. Stripes are crisp but their edges stair-step at the body cell; nose leather, eyes and claws separate by hue only, with no roughness split. No whiskers.", "fixes": [ - "fur cards or shells along the hair flow (planned M4 creature kind)", - "a fur material strip that does not darken the vertex colour (the library `fur` strip's brown albedo mutes orange, so plain was used)", - "per-vertex stripe colour on a finer head/body cell at LOD0 or a stripe texture lane" + "fur shells or cards along the hair flow on the back, ruff and tail (fur.ts) so the coat breaks light like fur", + "roughness split: wet nose leather and eyes glossy, coat fully rough", + "stripe colour on a finer head/body cell at LOD0 so stripe edges stop stair-stepping" ] }, "detail": { "score": 6, - "reason": "Primary and secondary forms are there (scapula, triceps, haunch, hock point, toe beans split by fur partings, sheathed claw tips, lidded eyes with round pupils, whisker-pad dots, ear ocelli, canines rooted in the gum, belly fringe, ruff tufts) and a strand relief at LOD0, but in a close-up there are no whiskers, no fur locks breaking the silhouette, and the face planes are soft.", + "reason": "Primary and secondary forms are there (scapula, haunch, hock point, toe beans, sheathed claws, lidded eyes, whisker-pad dots, ear ocelli, canines, belly fringe, ruff tufts), but in the portrait close-up there are no whiskers, no fur locks breaking the outline, and the face planes are soft.", "fixes": [ - "whisker strands as thin tubes from the pads", - "fur locks on the ruff/elbows/belly as cards", + "whisker strands as thin tubes from the pads (face.ts)", + "fur locks on the ruff, elbows and belly as cards", "sharper cheekbone and brow planes in face.ts" ] }, "wear": { "score": 4, - "reason": "A clean zoo animal: no mud on the paws, no scars, no matted or wet fur, no dust; only a ground-occlusion grade and a value gradient feet to head.", + "reason": "A clean zoo animal: no mud on the paws, no scars, no matted or wet fur, no dust; only a ground-occlusion grade.", "fixes": [ - "grime the paws and lower legs with soil colour (grime({ ground }))", - "a few scars on the face/shoulders and a torn ear notch, seeded", - "damp darker fur on the lower legs" + "grime the paws and lower legs with soil colour (grime with the ground mask)", + "a few seeded scars on the face/shoulders and a notch in one ear", + "damp, darker fur on the lower legs and belly fringe" ] }, "lod": { - "score": 7, - "reason": "LOD2 (2,900 tris) keeps the outline (IoU 0.93-0.95) and the orange/white blocks, but its legs are visibly tubes, the stripes become a blotchy pattern at the 7.8 cm cell, and the eyes are dropped (painted socket only).", + "score": 6, + "reason": "LOD2 (2.9k tris) keeps the outline and the orange/white blocks, but the far view in sheet.png is visibly faceted, the legs are tubes and the stripes become random dark triangles, so the pattern shifts at the switch.", "fixes": [ - "paint LOD2 stripes as 6-8 broad bands instead of sampling the LOD0 pattern", - "use the sculpt for LOD2 lower legs with a gentle inflate instead of tubes" + "paint LOD2 stripes as 6–8 broad authored bands instead of sampling the LOD0 pattern per triangle", + "build LOD2 lower legs from the sculpt with a gentle inflate instead of tubes" ] }, "technical": { "score": 8, - "reason": "QA PASS at every LOD including the new connections gate (anatomical folds declared as creases, each with a reason); 136.9k / 27.2k / 2.9k triangles under the 150k / 30k / 3k budgets; 3 draws; one material; compile 7.9 s / 1.6 s / 0.2 s; deterministic hash-seeded stripes; mass gate on a stated furred density.", + "reason": "QA PASS at every LOD; 136.9k / 27.2k / 2.9k triangles under the 150k / 30k / 3k budgets; 3 draws; one material; deterministic stripes.", "fixes": [ - "spend the claws' and ears' LOD0 triangles on the face instead (ears 5.5k each is more than they show)", - "merge eyes into one part per side with the lids for one fewer node pair" + "move LOD0 triangles from the ears and claws (dense for what they show) to the face", + "merge each eye with its lids into one part per side" ] } }, "renders": { - "hero.png": "1bc5dbcfafeac2f42fb94acb9ed91284103e0fef24a93da6c930c450e4b5619d", - "sheet.png": "0f35bea1ae933e5f16e17e831fa8ccebc2a9c36adaf0bfd26e8a009d5a491728", - "raking.png": "3def99ab6a44bed0ea33e66aaaef657596e6eeb914489940023e0b218c069767", - "raking_detail.png": "76fc04d2372ce7fc7f5a4275ef0292ddcb135a48069492061b7cb3449746ba8b", - "lods.png": "dbc433deaba4a8477e1ba36f6828d0c686ce1a83e0b4292793238d1467fc1c4b", - "silhouette.png": "95fd971163c9863ff2edb8b1d905df100226d6fc0335c45a93fd0c8c1ac60270", - "value.png": "415270e0748761c67298d01f0ba502db36a036218b7abab564efb398708a3558", - "detail.png": "6db9db51e9869dd066517a620852cee73c089d87c0dda3a4a90e060c2923e1a3" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.946 - }, - { - "view": "side", - "iou": 0.908 - }, - { - "view": "3/4", - "iou": 0.949 - } - ], - "focal": { - "peakPart": "head", - "peak": [ - 480, - 206 - ], - "focalShare": 0.779, - "focalArea": 0.231, - "top": [ - { - "part": "head", - "share": 0.779 - }, - { - "part": "body", - "share": 0.158 - }, - { - "part": "jaw", - "share": 0.029 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.566, - "spread": 0.541, - "min": 0.25 - }, - "grounded": { - "rise": 0.118, - "bands": [ - 0.415, - 0.337, - 0.357, - 0.446, - 0.479 - ] - }, - "shape": { - "acuteShare": 0.076, - "straightShare": 0.496, - "taper": -0.503, - "shape": "friendly", - "declared": "friendly" - } + "hero.png": "9c707fe4bb4dd30aab1c9d970778d5e284d7d967bd50236b73c4157b5adfb3ea", + "sheet.png": "d5d5027ec12f70e220ae851dc9a3351c8d588726f913e66682011dc87026230a" } } diff --git a/recipes/asteroid/out/asteroid/critique.json b/recipes/asteroid/out/asteroid/critique.json index 85c37c0e..febbb7f4 100644 --- a/recipes/asteroid/out/asteroid/critique.json +++ b/recipes/asteroid/out/asteroid/critique.json @@ -1,105 +1,65 @@ { "asset": "asteroid", "reviewer": "agent", - "overall": 7.4, + "overall": 6.7, "scores": { "silhouette": { - "score": 8, - "reason": "Lumpy carbonaceous asteroid with two big craters, smaller pits and a cluster of glowing crystals at mining sites; distinct per seed.", - "fixes": [] + "score": 7, + "reason": "Lumpy rock with craters, pits and glowing crystal clusters, distinct per seed. But the base is clipped flat at z = 0 and the fracture half-spaces cut near-vertical walls, so FRONT and SIDE in sheet.png read as a rounded box on a floor, not a free body in space.", + "fixes": [ + "drop the flat z = 0 clip (bounds from below the body) or keep it only for a ground-resting variant", + "raise the fracture-plane reach (rng.range(0.62, 0.8) → 0.75–0.9) and use fewer chips so they facet the lumps instead of squaring the sides" + ] }, "proportions": { "score": 8, - "reason": "Declared envelope per variant; crater sizes in proportion to the body.", - "fixes": [] + "reason": "Declared envelope per variant; crater sizes in proportion to the body; crystals at a believable scale against the 0.5 m grid.", + "fixes": [ + "vary the aspect between variants more (one elongated ~2:1 body) as real small bodies are rarely this blocky" + ] }, "materials": { - "score": 7, - "reason": "Regolith on the asphalt strip (dark aggregate, hairline cracks, rough) tinted to carbonaceous values, ochre ore veins, cyan crystal veins near the mining sites, glowing crystals. The aggregate is fine-grained for a 10 m body; no metallic flecks.", + "score": 6, + "reason": "Regolith on the asphalt strip tinted carbonaceous, ochre ore veins, cyan glowing crystals. The asphalt pattern reads as a fine tiled tread in hero.png, the hero is close to black with little value range, and the vein edges stair-step at the vertex cell in every sheet view.", "fixes": [ - "a project regolith material with larger clasts and metallic (iron-nickel) flecks" + "a project regolith material with larger clasts and iron-nickel flecks instead of the asphalt strip", + "blend the vein colour across vertices with a falloff (or sample it per pixel) so veins stop stair-stepping", + "lift the regolith albedo one step so craters read in the hero" ] }, "detail": { - "score": 7, - "reason": "Craters with rims, pits, veins, crystal clusters; SDF surface at 0.17 m.", + "score": 6, + "reason": "Craters with rims, pits, veins and crystal clusters; between them the surface is smooth blobby SDF at 0.17 m with no rubble or boulders.", "fixes": [ - "boulders resting in the crater floors" + "boulders and rubble resting in crater floors", + "a finer secondary noise octave at LOD0 for rock grain" ] }, "wear": { - "score": 7, - "reason": "Dust settled in up-facing hollows, darkened crater floors; space weathering otherwise uniform.", + "score": 5, + "reason": "Darkened crater floors and veins; otherwise uniform. No fresh ejecta, no mining scars around the crystal sites.", "fixes": [ - "bright fresh ejecta rays around the youngest crater" + "bright fresh ejecta rays around the youngest crater", + "drill scars and tailings around the crystal mining sites" ] }, "lod": { "score": 7, - "reason": "LOD2 (\u2248 600) keeps the outline, craters as colour, and the crystals.", - "fixes": [] + "reason": "LOD2 (636) keeps the outline, the crystals and the colour blocks, but the veins become large triangle patches.", + "fixes": [ + "drop the vein colour at LOD2 (keep the base tint) so it does not turn into triangle patches" + ] }, "technical": { "score": 8, - "reason": "QA PASS all variants: \u2248 20k / 4.4k / 0.6k, glow on the crystals, two draws.", - "fixes": [] + "reason": "QA PASS; 20 164 / 4 260 / 636 triangles, two draws, glow on the crystals.", + "fixes": [ + "shrink the sdf bounds to the body extent (ext = max size + 3.2 m wastes cells) to cut build time" + ] } }, "renders": { - "hero.png": "9a255b253297cc13066f78de98e24af1c87ee2d629876f1a2d4bfcdc6dfbc4e4", - "sheet.png": "aba7e46fce33a10297bbf82ee5409493743715f3d16ed7491d5e76f2fe533b5e", - "raking.png": "7c2992576fcfde92649da9a26646c45d01a6255f10d9e8b6de5dac0e9e9e89e2", - "raking_detail.png": "b7ad0aed8fdeaaed03e09eb94561a9badefce67eecb7893dcfcffc7db9e38f6d", - "lods.png": "6fcbde6f883b26f492e510f310bc6a3770011375592e260f4c35f9e8836de19e", - "silhouette.png": "9b43f9a0f322940a6e19c62d8712cecc4a1618f3ebcb8c6c8b59d3bb0baac34e", - "value.png": "c6be19e0939cc2f25f9f32e3be9109a9a9ffd2006fcb61e9886b32e877f74748", - "detail.png": "6a432a60739006d194bf4e23c366a5bb4365c956de777347539f87a611c2f00a" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.93 - }, - { - "view": "side", - "iou": 0.924 - }, - { - "view": "3/4", - "iou": 0.918 - } - ], - "focal": { - "peakPart": "ore_crystals", - "peak": [ - 215, - 178 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "ore_crystals", - "share": 0.669 - }, - { - "part": "rock", - "share": 0.331 - } - ], - "declared": [] - }, - "rest": { - "share": 0.799, - "spread": 0.462, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.208, - "straightShare": 0.505, - "taper": 0.188, - "shape": "stable" - } + "hero.png": "e3fe3dab046497db4745b45857f7336f4bf9b9f5ae627c046f24763a96d20314", + "sheet.png": "5e970d3f31cebeb79c6d8225f9dd2749ae464bc2b99488b80661fb084a487309" } } diff --git a/recipes/asteroid/out/asteroid_b/critique.json b/recipes/asteroid/out/asteroid_b/critique.json index 520f6045..b4ef6634 100644 --- a/recipes/asteroid/out/asteroid_b/critique.json +++ b/recipes/asteroid/out/asteroid_b/critique.json @@ -1,105 +1,65 @@ { "asset": "asteroid_b", "reviewer": "agent", - "overall": 7.4, + "overall": 6.7, "scores": { "silhouette": { - "score": 8, - "reason": "Lumpy carbonaceous asteroid with two big craters, smaller pits and a cluster of glowing crystals at mining sites; distinct per seed.", - "fixes": [] + "score": 7, + "reason": "Lumpy rock with craters, pits and glowing crystal clusters, distinct per seed. But the base is clipped flat at z = 0 and the fracture half-spaces cut near-vertical walls, so FRONT and SIDE in sheet.png read as a rounded box on a floor, not a free body in space.", + "fixes": [ + "drop the flat z = 0 clip (bounds from below the body) or keep it only for a ground-resting variant", + "raise the fracture-plane reach (rng.range(0.62, 0.8) → 0.75–0.9) and use fewer chips so they facet the lumps instead of squaring the sides" + ] }, "proportions": { "score": 8, - "reason": "Declared envelope per variant; crater sizes in proportion to the body.", - "fixes": [] + "reason": "Declared envelope per variant; crater sizes in proportion to the body; crystals at a believable scale against the 0.5 m grid.", + "fixes": [ + "vary the aspect between variants more (one elongated ~2:1 body) as real small bodies are rarely this blocky" + ] }, "materials": { - "score": 7, - "reason": "Regolith on the asphalt strip (dark aggregate, hairline cracks, rough) tinted to carbonaceous values, ochre ore veins, cyan crystal veins near the mining sites, glowing crystals. The aggregate is fine-grained for a 10 m body; no metallic flecks.", + "score": 6, + "reason": "Regolith on the asphalt strip tinted carbonaceous, ochre ore veins, cyan glowing crystals. The asphalt pattern reads as a fine tiled tread in hero.png, the hero is close to black with little value range, and the vein edges stair-step at the vertex cell in every sheet view.", "fixes": [ - "a project regolith material with larger clasts and metallic (iron-nickel) flecks" + "a project regolith material with larger clasts and iron-nickel flecks instead of the asphalt strip", + "blend the vein colour across vertices with a falloff (or sample it per pixel) so veins stop stair-stepping", + "lift the regolith albedo one step so craters read in the hero" ] }, "detail": { - "score": 7, - "reason": "Craters with rims, pits, veins, crystal clusters; SDF surface at 0.17 m.", + "score": 6, + "reason": "Craters with rims, pits, veins and crystal clusters; between them the surface is smooth blobby SDF at 0.17 m with no rubble or boulders.", "fixes": [ - "boulders resting in the crater floors" + "boulders and rubble resting in crater floors", + "a finer secondary noise octave at LOD0 for rock grain" ] }, "wear": { - "score": 7, - "reason": "Dust settled in up-facing hollows, darkened crater floors; space weathering otherwise uniform.", + "score": 5, + "reason": "Darkened crater floors and veins; otherwise uniform. No fresh ejecta, no mining scars around the crystal sites.", "fixes": [ - "bright fresh ejecta rays around the youngest crater" + "bright fresh ejecta rays around the youngest crater", + "drill scars and tailings around the crystal mining sites" ] }, "lod": { "score": 7, - "reason": "LOD2 (\u2248 600) keeps the outline, craters as colour, and the crystals.", - "fixes": [] + "reason": "LOD2 (548) keeps the outline, the crystals and the colour blocks, but the veins become large triangle patches.", + "fixes": [ + "drop the vein colour at LOD2 (keep the base tint) so it does not turn into triangle patches" + ] }, "technical": { "score": 8, - "reason": "QA PASS all variants: \u2248 20k / 4.4k / 0.6k, glow on the crystals, two draws.", - "fixes": [] + "reason": "QA PASS; 18 044 / 3 920 / 548 triangles, two draws, glow on the crystals.", + "fixes": [ + "shrink the sdf bounds to the body extent (ext = max size + 3.2 m wastes cells) to cut build time" + ] } }, "renders": { - "hero.png": "ca13b3f678c3806edaa9e52e6c4c7eeb79ee2df7d51d13254d91519f4d7a0e9d", - "sheet.png": "973a66474cd854ffebf04a330facb6f1d07c9f9867e3db6ff61c61456e227c10", - "raking.png": "7a0fbe5aa9213d61c29f191a8d512d967639a435a367b58dd97221c9b3b85887", - "raking_detail.png": "9bd638516b2842281bbb9219fab724c603088acdd4a18d7d51c9e7e38ebdf927", - "lods.png": "0cc0f5bb581e0797ebc5dcba802bf9f2674159f3279314e562a1a5d5b966d2cc", - "silhouette.png": "d9f977088c6c86fcbbac384f11a1ee12695ecf2ccb990eb8420e9ddee3d7cdfc", - "value.png": "ae8b013068b93170f4b0a70c3519cd0ae00a5c8f6a19778609128d76f62484f8", - "detail.png": "55db4908b53dfa383ebfea383a352b85d349bc698228c9a399f6deb397c8fcd0" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.884 - }, - { - "view": "side", - "iou": 0.898 - }, - { - "view": "3/4", - "iou": 0.875 - } - ], - "focal": { - "peakPart": "ore_crystals", - "peak": [ - 288, - 176 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "ore_crystals", - "share": 0.655 - }, - { - "part": "rock", - "share": 0.345 - } - ], - "declared": [] - }, - "rest": { - "share": 0.75, - "spread": 0.547, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.408, - "straightShare": 0.491, - "taper": 0.541, - "shape": "aggressive" - } + "hero.png": "0c1b61eb653e5a34175f3c4acad567da4a0ba649cc54ce211618d0356a913651", + "sheet.png": "9a12d0097ecea3894417c60800a3c909d0db939f73b4e3922776ccff73986ffd" } } diff --git a/recipes/asteroid/out/asteroid_c/critique.json b/recipes/asteroid/out/asteroid_c/critique.json index 9f8c1e30..c7971be1 100644 --- a/recipes/asteroid/out/asteroid_c/critique.json +++ b/recipes/asteroid/out/asteroid_c/critique.json @@ -1,105 +1,65 @@ { "asset": "asteroid_c", "reviewer": "agent", - "overall": 7.4, + "overall": 6.7, "scores": { "silhouette": { - "score": 8, - "reason": "Lumpy carbonaceous asteroid with two big craters, smaller pits and a cluster of glowing crystals at mining sites; distinct per seed.", - "fixes": [] + "score": 7, + "reason": "Lumpy rock with craters, pits and glowing crystal clusters, distinct per seed. But the base is clipped flat at z = 0 and the fracture half-spaces cut near-vertical walls, so FRONT and SIDE in sheet.png read as a rounded box on a floor, not a free body in space.", + "fixes": [ + "drop the flat z = 0 clip (bounds from below the body) or keep it only for a ground-resting variant", + "raise the fracture-plane reach (rng.range(0.62, 0.8) → 0.75–0.9) and use fewer chips so they facet the lumps instead of squaring the sides" + ] }, "proportions": { "score": 8, - "reason": "Declared envelope per variant; crater sizes in proportion to the body.", - "fixes": [] + "reason": "Declared envelope per variant; crater sizes in proportion to the body; crystals at a believable scale against the 0.5 m grid.", + "fixes": [ + "vary the aspect between variants more (one elongated ~2:1 body) as real small bodies are rarely this blocky" + ] }, "materials": { - "score": 7, - "reason": "Regolith on the asphalt strip (dark aggregate, hairline cracks, rough) tinted to carbonaceous values, ochre ore veins, cyan crystal veins near the mining sites, glowing crystals. The aggregate is fine-grained for a 10 m body; no metallic flecks.", + "score": 6, + "reason": "Regolith on the asphalt strip tinted carbonaceous, ochre ore veins, cyan glowing crystals. The asphalt pattern reads as a fine tiled tread in hero.png, the hero is close to black with little value range, and the vein edges stair-step at the vertex cell in every sheet view.", "fixes": [ - "a project regolith material with larger clasts and metallic (iron-nickel) flecks" + "a project regolith material with larger clasts and iron-nickel flecks instead of the asphalt strip", + "blend the vein colour across vertices with a falloff (or sample it per pixel) so veins stop stair-stepping", + "lift the regolith albedo one step so craters read in the hero" ] }, "detail": { - "score": 7, - "reason": "Craters with rims, pits, veins, crystal clusters; SDF surface at 0.17 m.", + "score": 6, + "reason": "Craters with rims, pits, veins and crystal clusters; between them the surface is smooth blobby SDF at 0.17 m with no rubble or boulders.", "fixes": [ - "boulders resting in the crater floors" + "boulders and rubble resting in crater floors", + "a finer secondary noise octave at LOD0 for rock grain" ] }, "wear": { - "score": 7, - "reason": "Dust settled in up-facing hollows, darkened crater floors; space weathering otherwise uniform.", + "score": 5, + "reason": "Darkened crater floors and veins; otherwise uniform. No fresh ejecta, no mining scars around the crystal sites.", "fixes": [ - "bright fresh ejecta rays around the youngest crater" + "bright fresh ejecta rays around the youngest crater", + "drill scars and tailings around the crystal mining sites" ] }, "lod": { "score": 7, - "reason": "LOD2 (\u2248 600) keeps the outline, craters as colour, and the crystals.", - "fixes": [] + "reason": "LOD2 (616) keeps the outline, the crystals and the colour blocks, but the veins become large triangle patches.", + "fixes": [ + "drop the vein colour at LOD2 (keep the base tint) so it does not turn into triangle patches" + ] }, "technical": { "score": 8, - "reason": "QA PASS all variants: \u2248 20k / 4.4k / 0.6k, glow on the crystals, two draws.", - "fixes": [] + "reason": "QA PASS; 20 392 / 4 324 / 616 triangles, two draws, glow on the crystals.", + "fixes": [ + "shrink the sdf bounds to the body extent (ext = max size + 3.2 m wastes cells) to cut build time" + ] } }, "renders": { - "hero.png": "66812367225d9b0c7ddc17efb5a5e775c211d20cb35856db8c803af559e9bc9a", - "sheet.png": "c02ab89d27b01449a56aa2de174c764960793c0ca6b82101cff07f0e7ede6546", - "raking.png": "d34683a607a29031259459331e28aa3de54ddc0b1639e028a0c1c9e02df88705", - "raking_detail.png": "0bc6eb206f4f4cc7cbb9d0c85204d7fbee7a0fc0cf78e543c551b96e6104bf31", - "lods.png": "d594e6e8e00a18baf64043cf953a46a70ec29b9a9eb3a5399f2e0b6e5db28fb6", - "silhouette.png": "f98db5404cc38ecdc070a325daa375c555dc4edcb7b9816044789fd37539cd28", - "value.png": "b100cb893fda2635efb37cc556f7617c9cc1f57feee14a56a76434f3f70c0128", - "detail.png": "a18f97d21786ac95e2eda5394e08a80f3e6602a01803412fc7676151000f96ad" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.889 - }, - { - "view": "side", - "iou": 0.911 - }, - { - "view": "3/4", - "iou": 0.907 - } - ], - "focal": { - "peakPart": "ore_crystals", - "peak": [ - 257, - 163 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "ore_crystals", - "share": 0.705 - }, - { - "part": "rock", - "share": 0.295 - } - ], - "declared": [] - }, - "rest": { - "share": 0.669, - "spread": 0.62, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.325, - "straightShare": 0.433, - "taper": 0.59, - "shape": "aggressive" - } + "hero.png": "4cea3b669950e5c0729873aad0186d84dc7b5ac0294cbddb4e5360d66da4bcd4", + "sheet.png": "e87df3b88c1dd0fa5da673126fd59f136051f3cd4ffbfeda350b13d33b50c260" } } diff --git a/recipes/bear_hybrid/README.md b/recipes/bear_hybrid/README.md index 8cef9787..03969edf 100644 --- a/recipes/bear_hybrid/README.md +++ b/recipes/bear_hybrid/README.md @@ -36,6 +36,12 @@ This recipe declares the animal the model shows, inside the same sourced ranges. Five cached requests, each naming the previous output by sha256 (kinds `modelChain`). A build never calls a model or Blender; `t3d neural run` does. +The chain is `required: true`. The code bear this spec would give without the model is not this +asset (over 100 000 triangles at LOD 0 against a 30 000 budget, off the sourced box), so `t3d build` +with an output missing stops with one `T3D_NEURAL_MISSING` and builds nothing. The npm package does +not ship `neural/`: clone the repository to build this recipe, or copy the folder into your project +and run `t3d neural run` on the copy (GPU and Blender, below). + | # | Task | Tool (licence) | Output | In git | Time (RTX 3090) | |---|---|---|---|---|---| | 1 | `image` | Z-Image-Turbo (Apache-2.0), seed 5, 1024² | concept PNG, 1.0 MB | yes | 38 s | diff --git a/recipes/bear_hybrid/bear_hybrid.test.ts b/recipes/bear_hybrid/bear_hybrid.test.ts index 8b43f35c..e8f9f31f 100644 --- a/recipes/bear_hybrid/bear_hybrid.test.ts +++ b/recipes/bear_hybrid/bear_hybrid.test.ts @@ -7,7 +7,14 @@ import { afterEach, describe, expect, it } from 'bun:test'; import { join } from 'node:path'; -import { clearNeuralSources, compile, runQa, toGlb } from '@developerz.ai/ts23d'; +import { + clearNeuralSources, + compile, + neuralCache, + provideNeural, + runQa, + toGlb, +} from '@developerz.ai/ts23d'; import { compileLods } from '@developerz.ai/ts23d/testing'; import { provideRecipeNeural } from '../../src/cli/neural-store.ts'; import { bear_hybrid, measured, SOURCED } from './bear_hybrid.ts'; @@ -39,6 +46,12 @@ describe('bear_hybrid, CPU tier (no neural cache)', () => { }); describe('bear_hybrid, model tier (committed cache)', () => { + it('a build with an empty cache (the npm package) stops with T3D_NEURAL_MISSING, no CPU stand-in', () => { + clearNeuralSources(); + provideNeural(neuralCache({ recipe: 'bear_hybrid', offline: true })); + expect(() => compile(bear_hybrid, { lod: 2 })).toThrow('T3D_NEURAL_MISSING'); + }); + it( 'a missing cache is not mistaken for the model: the textured body appears only with it', async () => { diff --git a/recipes/bear_hybrid/out/bear_hybrid/critique.json b/recipes/bear_hybrid/out/bear_hybrid/critique.json new file mode 100644 index 00000000..a9f1a83c --- /dev/null +++ b/recipes/bear_hybrid/out/bear_hybrid/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "bear_hybrid", + "reviewer": "agent", + "overall": 7.4, + "scores": { + "silhouette": { + "score": 8, + "reason": "Unmistakably a brown bear from every view: shoulder hump, dished face, round ears, heavy plantigrade forelegs, short tail; the side view in sheet.png matches the classic standing-bear outline.", + "fixes": [ + "raise the shoulder hump 3–4 cm in the concept/mesh chain so it reads more sharply against the back line in SIDE" + ] + }, + "proportions": { + "score": 8, + "reason": "1.7 × 0.6 × 1.0 m on the grid matches the sourced average adult (1.7 m head and body, 1.0 m shoulder). Head-to-body and leg lengths look right beside photos; the ears are slightly large.", + "fixes": [ + "scale the ears ~15% down in a Blender step (steps.ts) — they read cub-sized against the head" + ] + }, + "materials": { + "score": 7, + "reason": "The image-to-3D texture reads as fur with darker legs and a grizzled back; claws separate. But lighting is baked into the albedo (a bright shoulder patch, white-capped ears that look sunlit from every angle) and the fur has one uniform roughness.", + "fixes": [ + "de-light the albedo before import (a Blender bake step that divides out the baked AO/sun)", + "repaint the ear tips to the body brown; they read as white caps" + ] + }, + "detail": { + "score": 7, + "reason": "Fur breakup, face features and claws come from the texture and hold at third-person distance; geometry is smooth (no fur clumps on the outline, claws painted rather than modelled).", + "fixes": [ + "a strand/shell pass on the back and belly fringe so the outline breaks into fur clumps", + "model claws as short curved cones on each paw (code parts on the paw bones)" + ] + }, + "wear": { + "score": 6, + "reason": "Darker, dirtier legs and a grizzled back give some story, but there is no mud, no wet fur, no scars or rubbed patches.", + "fixes": [ + "mud on the paws and lower legs (grime with the ground mask over the textured part)", + "a few seeded scars and a rubbed patch on the shoulder" + ] + }, + "lod": { + "score": 8, + "reason": "LOD2 (5k) is nearly indistinguishable from LOD0 at the sheet's far distance: same outline and colour blocks.", + "fixes": [ + "drop LOD2 to ~3k: the far view does not need 5k for a 1.7 m animal, which would pay for detail at LOD0" + ] + }, + "technical": { + "score": 8, + "reason": "QA PASS; 30k / 10k / 5k triangles; one draw per LOD; skinned to the sourced skeleton with idle, look and walk clips.", + "fixes": [ + "spend the LOD2 savings on the face at LOD0 (eye sockets, lips) where the hero looks" + ] + } + }, + "renders": { + "hero.png": "390997cbafdde821d60b6ff7ae4324d9d6518d2f88a64854c99a6eee0122d995", + "sheet.png": "7fa1c5744f374d332b1c9ac9176d36b99bc6b0be325fc114d2b4542819d8256c" + } +} diff --git a/recipes/bear_hybrid/surface.ts b/recipes/bear_hybrid/surface.ts index d1261d12..2a56d2c2 100644 --- a/recipes/bear_hybrid/surface.ts +++ b/recipes/bear_hybrid/surface.ts @@ -9,7 +9,10 @@ // elbows and the spine lie (mapped by topology) // blender FINAL: transform(fit) + steps.ts → the game mesh: three LODs, PNG maps, the skin; // committed, the only thing a build draws -// A miss anywhere returns null: the code bear (bear_hybrid.ts). Nothing here calls a model; +// The chain is `required`: the code creature of this spec fails QA (over 100 000 triangles at +// LOD 0, off the sourced box), so `t3d build` with an output missing stops with one +// T3D_NEURAL_MISSING and never ships it. A compile with no cache provided at all (a test, the +// gallery's CPU tier) still gets null: the code bear (bear_hybrid.ts). Nothing here calls a model; // `t3d neural run` does, on a machine with a GPU and Blender (README.md). import { @@ -39,7 +42,9 @@ export const SURFACE: Omit = { * plantigrade wrist 84° back, and a thick furred paw passes through the forearm there. */ export const DROPPED: readonly string[] = ['trot', 'gallop']; -/** The imported game mesh with the rig it was built for, or null on any cache miss. */ +/** The imported game mesh with the rig it was built for; null with no cache provided + * and while `t3d neural run` is still filling the cache; an offline (build) cache throws + * `T3D_NEURAL_MISSING` on a miss instead. */ export const bearModel = (ctx: BuildContext): ChainedModel | null => modelChain(ctx, { concept: CONCEPT, @@ -53,6 +58,7 @@ export const bearModel = (ctx: BuildContext): ChainedModel | null => maxTriangles: TRIANGLES[0], models: { image: CONCEPT_MODEL, mesh: SURFACE_MODEL }, license: 'MIT', + required: true, author: 'ts23d bear_hybrid recipe (Z-Image concept, Pixal3D mesh, UniRig joints, Blender steps)', }); diff --git a/recipes/bluefin_tuna/out/bluefin_tuna/critique.json b/recipes/bluefin_tuna/out/bluefin_tuna/critique.json new file mode 100644 index 00000000..a8bfe3c5 --- /dev/null +++ b/recipes/bluefin_tuna/out/bluefin_tuna/critique.json @@ -0,0 +1,64 @@ +{ + "asset": "bluefin_tuna", + "reviewer": "agent", + "overall": 7, + "scores": { + "silhouette": { + "score": 8, + "reason": "Fusiform body, tall sickle second dorsal and anal fins, yellow finlet rows, short pectorals and a crescent tail: names a bluefin tuna in SIDE and TOP at once.", + "fixes": [ + "make the caudal keels visible as a ridge on the tail stock in SIDE/TOP (fins.ts)" + ] + }, + "proportions": { + "score": 8, + "reason": "2.1 × 0.6 × 1.1 m for a 2 m fork length; body depth, fin positions and finlet count look right beside the dossier photos.", + "fixes": [ + "check the pectoral length against a side photo (it reads slightly long for Thunnus thynnus)" + ] + }, + "materials": { + "score": 7, + "reason": "Metallic blue back with a strong highlight over a silver flank, yellow finlets and fin bases; the silver flank reads as flat grey paint rather than mirror-like scales.", + "fixes": [ + "raise metalness / lower roughness on the flank so it reflects like a fish side", + "iridescent band and faint silver spots/bars on the lower flank" + ] + }, + "detail": { + "score": 6, + "reason": "Operculum line, eye with ring, finlets, lateral bars; but the body is a smooth shell (no lateral line, no corselet scale patch) and the fins are flat plates with no rays.", + "fixes": [ + "fin rays as shallow grooves on the dorsal, anal and caudal fins (fins.ts)", + "lateral line and corselet scale patch behind the head", + "gill slit depth at the operculum edge" + ] + }, + "wear": { + "score": 5, + "reason": "Pristine: no frayed fin edges, no scratches, no slime sheen variation.", + "fixes": [ + "slightly frayed fin margins and a few scratches on the flank (seeded)" + ] + }, + "lod": { + "score": 7, + "reason": "LOD2 (476 tris) keeps the outline, fins and the blue-over-silver split, but the body is clearly faceted and the finlets thin out.", + "fixes": [ + "keep finlets at LOD2 as a single flat yellow strip per side", + "give LOD2 more rings around the body (~700 tris) so the faceting stops reading" + ] + }, + "technical": { + "score": 8, + "reason": "QA PASS; 32k / 4k / 476 triangles, one draw.", + "fixes": [ + "LOD1 at 4k drops a lot from 32k: add an intermediate ring count so the LOD0→LOD1 switch is softer" + ] + } + }, + "renders": { + "hero.png": "2f0b8329de398ba4750a6c207b6b63126ac61e98b998c372f19a702b04e5fdd4", + "sheet.png": "0133bf0e8f82acdbfbbe6c065824ad981901bd5ede3b72f76365749ab786d11b" + } +} diff --git a/recipes/boulders/out/boulders/critique.json b/recipes/boulders/out/boulders/critique.json index e85012aa..157a440a 100644 --- a/recipes/boulders/out/boulders/critique.json +++ b/recipes/boulders/out/boulders/critique.json @@ -1,23 +1,28 @@ { "asset": "boulders", "reviewer": "agent", - "overall": 7.6, + "overall": 7.3, "scores": { "silhouette": { "score": 8, - "reason": "Four boulders 0.6\u20132.4 m with fracture-cut faces and rounded weathered masses; each reads as a different rock.", - "fixes": [] + "reason": "Four boulders 0.6–2.4 m with fracture-cut faces and rounded weathered masses; each reads as a different rock in the row.", + "fixes": [ + "make one boulder flatter and slab-like so the set has a third shape family (round, angular, slab)" + ] }, "proportions": { "score": 8, - "reason": "Sizes as declared; flattened, ground-seated bases.", - "fixes": [] + "reason": "Sizes as declared on the 0.5 m grid; flattened, ground-seated bases.", + "fixes": [ + "sink the bases 5–10 cm so they read embedded rather than placed" + ] }, "materials": { - "score": 7, - "reason": "Granite cliff strip (joints, crystalline grain, normal relief) with a different variant per rock, darker damp feet and lighter tops. Thin dark contour outlines appear where the strip's lichen layer meets the projection seams.", + "score": 6, + "reason": "Granite cliff strip with joints and crystalline grain. In hero.png the third boulder shows jagged camouflage-like patches with hard pixel edges and the dark joint network reads as painted outlines; the triplanar seams show.", "fixes": [ - "blend the triplanar seams on SDF meshes (or a dedicated boulder material without the lichen layer)" + "blend the triplanar projection seams on the SDF meshes", + "a dedicated boulder material without the lichen/joint layer that draws the outlines" ] }, "detail": { @@ -28,82 +33,29 @@ ] }, "wear": { - "score": 8, - "reason": "Moss on the shaded north faces and the damp feet, pale lichen rosettes on sunny tops, darker bases: tells which side faces north and where water sits.", + "score": 7, + "reason": "Moss on the shaded faces and pale lichen on tops show in hero.png, but faintly; nothing at the base ties them to the ground.", "fixes": [ - "a few half-buried pebbles and soil at the base" + "stronger moss on the north faces and a few half-buried pebbles and soil at each base" ] }, "lod": { "score": 7, - "reason": "LOD2 (956) keeps all four masses; small fracture faces round off.", - "fixes": [] + "reason": "LOD2 (956) keeps all four masses; small fracture faces round off and the faceting shows.", + "fixes": [ + "keep the largest fracture plane per boulder at LOD2 (smax with a coarser cell) so the angular rocks stay angular" + ] }, "technical": { "score": 8, "reason": "QA PASS, 36 260 / 6 684 / 956, one draw.", - "fixes": [] + "fixes": [ + "drop LOD0 resolution on the smallest boulder (it gets the same cell as the 2.4 m one)" + ] } }, "renders": { - "hero.png": "3174ec719250365cc0a1e304b5b35249f975fdfd528598685ee64c88332638c5", - "sheet.png": "26a0c318ce1a2885b73402c43efda82fa56af9c9164df695bdbf7fd72b66ac62", - "raking.png": "ee286b0ef63829748d7c11be9e86c2b6c739dc9c44d7f57b6c36fc29cf643359", - "raking_detail.png": "6207dd5571e072761c4aaef01f7fdf8a16300f2d04866a94d165fe3649b88148", - "lods.png": "603e089b1c47494b94d39555565636efed675ef0f3df85de919b7390359179b8", - "silhouette.png": "93946db76dbe454174671ba16017976f71f8c15213b5b6d486f77ce98891ef19", - "value.png": "2f97b3b1855342e8315d58fd56bee4e2c0695f707c035637e4dee4726396aef7", - "detail.png": "bdd3b66f5065ffc77d487a179bae1b60a05095974a1c8584fdf3847623834bed" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.842 - }, - { - "view": "side", - "iou": 0.857 - }, - { - "view": "3/4", - "iou": 0.82 - } - ], - "focal": { - "peakPart": "boulder_3", - "peak": [ - 600, - 106 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "boulder_3", - "share": 0.423 - }, - { - "part": "boulder_2", - "share": 0.232 - }, - { - "part": "boulder_0", - "share": 0.193 - } - ], - "declared": [] - }, - "rest": { - "share": 0.323, - "spread": 0.544, - "min": 0.25 - }, - "shape": { - "acuteShare": 0, - "straightShare": 0.552, - "taper": 0.425, - "shape": "stable" - } + "hero.png": "fe4981a6534a647b8885506f0373367600619575fa834dcb528d6360d142e57e", + "sheet.png": "9011df587aac8619a96a2168f62172d231a6d5612c8d93437bd6325ecd5b07dd" } } diff --git a/recipes/boxcar/out/boxcar/critique.json b/recipes/boxcar/out/boxcar/critique.json index 91a70dbe..1c27a6db 100644 --- a/recipes/boxcar/out/boxcar/critique.json +++ b/recipes/boxcar/out/boxcar/critique.json @@ -1,122 +1,66 @@ { "asset": "boxcar", "reviewer": "agent", - "overall": 7.6, + "overall": 7.3, "scores": { "silhouette": { "score": 8, - "reason": "Single-sheathed USRA boxcar from every side: Howe-truss bracing, running board with laterals, sliding door on its track, Andrews-style trucks, brake staff and wheel at the B end. The underside now has brake gear: brake beams and shoes on every truck, the cylinder and floating levers with pull rods to the trucks, reservoir, triple valve, train line (LOD0 and LOD1).", + "reason": "Single-sheathed USRA boxcar from every side: Howe-truss bracing, running board with laterals, sliding door on its track, Andrews-style trucks, brake staff and wheel at the B end.", "fixes": [ - "thicker brake staff and a spoked wheel so the B end reads in the FRONT view" + "thicker brake staff and a spoked brake wheel so the B end reads in the FRONT view" ] }, "proportions": { "score": 8, - "reason": "13.4 \u00d7 2.9 \u00d7 4.04 m; 41 ft 3 in body, 33 in wheels, 5 ft 6 in wheelbase, door width and wheel spacing pass the \u00b12% measures; panel count each side of the door matches the Bettendorf photo.", + "reason": "13.4 × 3.0 × 4.1 m; body length, wheel size, wheelbase, door width and wheel spacing pass the ±2% measures; panel count each side of the door matches the builder photo.", "fixes": [ "check truck side-frame spacing against the FRONT view of the builder photo" ] }, "materials": { "score": 7, - "reason": "wood_painted siding (a different variant per side, same strip at every LOD so the red no longer shifts), steel_painted framing and ends, weathered-wood running board, cast-iron trucks and wheels, rusted hardware. Lettering now continues onto the diagonal braces, but at 3/4 the 6.5 cm brace offset still breaks 'CLINCHFIELD' by parallax. Wheel faces show radial lathe streaks.", + "reason": "Painted wood siding, steel framing and ends, weathered-wood running board, cast-iron trucks. The road name (CLINCHFIELD) is still broken by the diagonal braces' parallax at 3/4 in hero.png, and the running board renders near-white in sheet.png.", "fixes": [ - "planar-map the wheel faces instead of lathe UVs (cast-iron reads as brushed)", - "taper the brace face toward the letters or flush-mount stencil pieces" + "flush-mount the stencil pieces or taper the brace faces so the road name reads at 3/4", + "darken the running-board wood to a grey-brown weathered value", + "planar-map the wheel faces instead of lathe UVs" ] }, "detail": { "score": 7, - "reason": "Board-level siding, bolted bracing with rivets, door straps and handle, door track and guide, couplers with uncoupling lever and air hose, ladders and stirrups, brake beams and rigging. Still missing journal-box lids, door hasp/seal pin, roof seam caps.", + "reason": "Board-level siding, bolted bracing, door straps and handle, door track, couplers with uncoupling lever and air hose, ladders and stirrups, brake rigging. Missing journal-box lids, door hasp/seal pin, roof seam caps.", "fixes": [ "journal boxes with hinged lids", "door hasp, seal pin and door stops", - "end-roof fascia" + "roof seam caps" ] }, "wear": { - "score": 7, - "reason": "Road grime rises from the rails (darkest at the sill, fading by mid-height) with blotchy fade, roof darkened by locomotive smoke, rust streaks below the upper rivet rows and the door guide, rusted couplers. Streaks are small at hero distance; no replaced boards or chalk marks.", + "score": 6, + "reason": "The hero reads as a freshly painted car: the siding red is even, the grime gradient from the rails and the rust streaks under the rivets are too faint to see at hero distance; only the underframe is dark.", "fixes": [ - "a few replaced boards in fresher paint (board-aligned vertex colour)", - "chalk car-card marks by the door", + "stronger road grime rising from the sill (double the grime strength below 1 m)", + "a few replaced boards in fresher paint and chalk car-card marks by the door", "longer, wider rust streaks under the door track" ] }, "lod": { - "score": 8, - "reason": "LOD2 (1.2k) keeps the box, roof walk, trucks and the big white road name; same wood_painted strip at every LOD so the body red matches LOD0; bracing drops to texture.", + "score": 7, + "reason": "LOD2 (1.2k) keeps the box, roof walk, trucks and the road name, but drops all the Howe-truss bracing and the door outline, which are the biggest graphic marks on the side at mid distance.", "fixes": [ - "bake the brace pattern into LOD2 vertex colour" + "bake the brace pattern and the door outline into LOD2 vertex colour" ] }, "technical": { "score": 8, - "reason": "QA PASS both liveries: 25 628 / 9 272 / 1 192 (boxcar), 25 816 / 9 460 / 1 336 (boxcar_nw); the truck\u2194underframe slit is now declared on the family, so boxcar_nw passes too. One material; paint slots keep switchable liveries. Now 25 676 / 9 272 / 1 192. Connections gate clean (was 16 findings per livery): grab-iron legs bolt through the side boards / end wall (1 cm into them) instead of stopping 7\u20139 cm short at the post line; a train-line pipe carries the air hose out of the centre sill.", + "reason": "QA PASS; ~25.5k / ~9.2k / 1.2–1.3k triangles; one material; 13 / 11 / 11 draws for the moving pivots; switchable liveries via paint slots.", "fixes": [ "trim LOD0 by sharing one tessellated board panel per side" ] } }, "renders": { - "hero.png": "e525d4f8b51bd6b8fbfca66caed44fc54db33194e82df0b9ecdea4139713643e", - "sheet.png": "050c187e90f3900d953a0b4d457f5c646dfae8dff980ac3a1cb07fbb01f831c1", - "raking.png": "e105255e9d79e8b9a9459687fb32a8cb3185aa44c454c91a317a99a767f7718a", - "raking_detail.png": "8543a1551bc19a41558a6a960adffc64283b896e71fc841c9f40c4c3672b9c7f", - "lods.png": "0904a72e8c5b97724e91e04282a6017a1d7622853924cfb3131e0a6ab5b587d7", - "silhouette.png": "2b74b0d3e5268524dc10a9ffa8b9d079ed0bc0e7ab7249684918470d7f2f9514", - "value.png": "78d30a509a65265148ddbe2d5f030b12fb4bcc44adce69994757e6f9ea6b7cad", - "detail.png": "7fa40af4d47a8e369db89d681a433f9a6c2f6ff8fcc4632d04fe3701027e8cee" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.907 - }, - { - "view": "side", - "iou": 0.972 - }, - { - "view": "3/4", - "iou": 0.983 - } - ], - "focal": { - "peakPart": "body", - "peak": [ - 210, - 117 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "roof", - "share": 0.26 - }, - { - "part": "body", - "share": 0.245 - }, - { - "part": "body_framing", - "share": 0.18 - } - ], - "declared": [] - }, - "rest": { - "share": 0.412, - "spread": 0.781, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.213, - "straightShare": 0.77, - "taper": -0.314, - "shape": "stable" - } + "hero.png": "4ab3a7fa8625efc7a639657424f002136731edddb9d81273682c3d58064d93ea", + "sheet.png": "db1bb7ae555fac2aacb8818be7e70dcb67e38d87e7ee84e94b25c559e8d62bc1" } } diff --git a/recipes/boxcar/out/boxcar_nw/critique.json b/recipes/boxcar/out/boxcar_nw/critique.json index 13e8a9db..0d99efef 100644 --- a/recipes/boxcar/out/boxcar_nw/critique.json +++ b/recipes/boxcar/out/boxcar_nw/critique.json @@ -1,122 +1,66 @@ { "asset": "boxcar_nw", "reviewer": "agent", - "overall": 7.6, + "overall": 7.3, "scores": { "silhouette": { "score": 8, - "reason": "Single-sheathed USRA boxcar from every side: Howe-truss bracing, running board with laterals, sliding door on its track, Andrews-style trucks, brake staff and wheel at the B end. The underside now has brake gear: brake beams and shoes on every truck, the cylinder and floating levers with pull rods to the trucks, reservoir, triple valve, train line (LOD0 and LOD1).", + "reason": "Single-sheathed USRA boxcar from every side: Howe-truss bracing, running board with laterals, sliding door on its track, Andrews-style trucks, brake staff and wheel at the B end.", "fixes": [ - "thicker brake staff and a spoked wheel so the B end reads in the FRONT view" + "thicker brake staff and a spoked brake wheel so the B end reads in the FRONT view" ] }, "proportions": { "score": 8, - "reason": "13.4 \u00d7 2.9 \u00d7 4.04 m; 41 ft 3 in body, 33 in wheels, 5 ft 6 in wheelbase, door width and wheel spacing pass the \u00b12% measures; panel count each side of the door matches the Bettendorf photo.", + "reason": "13.4 × 3.0 × 4.1 m; body length, wheel size, wheelbase, door width and wheel spacing pass the ±2% measures; panel count each side of the door matches the builder photo.", "fixes": [ "check truck side-frame spacing against the FRONT view of the builder photo" ] }, "materials": { "score": 7, - "reason": "wood_painted siding (a different variant per side, same strip at every LOD so the red no longer shifts), steel_painted framing and ends, weathered-wood running board, cast-iron trucks and wheels, rusted hardware. Lettering now continues onto the diagonal braces, but at 3/4 the 6.5 cm brace offset still breaks 'CLINCHFIELD' by parallax. Wheel faces show radial lathe streaks.", + "reason": "Painted wood siding, steel framing and ends, weathered-wood running board, cast-iron trucks. The road name (NORFOLK & WESTERN) is still broken by the diagonal braces' parallax at 3/4 in hero.png, and the running board renders near-white in sheet.png.", "fixes": [ - "planar-map the wheel faces instead of lathe UVs (cast-iron reads as brushed)", - "taper the brace face toward the letters or flush-mount stencil pieces" + "flush-mount the stencil pieces or taper the brace faces so the road name reads at 3/4", + "darken the running-board wood to a grey-brown weathered value", + "planar-map the wheel faces instead of lathe UVs" ] }, "detail": { "score": 7, - "reason": "Board-level siding, bolted bracing with rivets, door straps and handle, door track and guide, couplers with uncoupling lever and air hose, ladders and stirrups, brake beams and rigging. Still missing journal-box lids, door hasp/seal pin, roof seam caps.", + "reason": "Board-level siding, bolted bracing, door straps and handle, door track, couplers with uncoupling lever and air hose, ladders and stirrups, brake rigging. Missing journal-box lids, door hasp/seal pin, roof seam caps.", "fixes": [ "journal boxes with hinged lids", "door hasp, seal pin and door stops", - "end-roof fascia" + "roof seam caps" ] }, "wear": { - "score": 7, - "reason": "Road grime rises from the rails (darkest at the sill, fading by mid-height) with blotchy fade, roof darkened by locomotive smoke, rust streaks below the upper rivet rows and the door guide, rusted couplers. Streaks are small at hero distance; no replaced boards or chalk marks.", + "score": 6, + "reason": "The hero reads as a freshly painted car: the siding red is even, the grime gradient from the rails and the rust streaks under the rivets are too faint to see at hero distance; only the underframe is dark.", "fixes": [ - "a few replaced boards in fresher paint (board-aligned vertex colour)", - "chalk car-card marks by the door", + "stronger road grime rising from the sill (double the grime strength below 1 m)", + "a few replaced boards in fresher paint and chalk car-card marks by the door", "longer, wider rust streaks under the door track" ] }, "lod": { - "score": 8, - "reason": "LOD2 (1.2k) keeps the box, roof walk, trucks and the big white road name; same wood_painted strip at every LOD so the body red matches LOD0; bracing drops to texture.", + "score": 7, + "reason": "LOD2 (1.3k) keeps the box, roof walk, trucks and the road name, but drops all the Howe-truss bracing and the door outline, which are the biggest graphic marks on the side at mid distance.", "fixes": [ - "bake the brace pattern into LOD2 vertex colour" + "bake the brace pattern and the door outline into LOD2 vertex colour" ] }, "technical": { "score": 8, - "reason": "QA PASS both liveries: 25 628 / 9 272 / 1 192 (boxcar), 25 816 / 9 460 / 1 336 (boxcar_nw); the truck\u2194underframe slit is now declared on the family, so boxcar_nw passes too. One material; paint slots keep switchable liveries. Now 25 864 / 9 460 / 1 336. Connections gate clean (was 16 findings): grab-iron legs bolt through the boards / end wall; train-line pipe holds the air hose.", + "reason": "QA PASS; ~25.5k / ~9.2k / 1.2–1.3k triangles; one material; 13 / 11 / 11 draws for the moving pivots; switchable liveries via paint slots.", "fixes": [ "trim LOD0 by sharing one tessellated board panel per side" ] } }, "renders": { - "hero.png": "1c20341aef27625100c2d17b485c7c7362b886de57d92a5ba48c4f2c583bfd77", - "sheet.png": "e70f015afd036085175c7102c84d2097e39c75ecab0c405a9b0728d5478e60b7", - "raking.png": "eaf5feb822c9f56df0e591a57d429f6e612d75262f74b60bcbf8777c37eb4b08", - "raking_detail.png": "8d99e8ad95bb9b98b758d93d1d16d4a66fab8efe06ae7b1e529103e54fd7cf5e", - "lods.png": "171fade1c7b169e48812ffed666cde968c242121ab6e1263630a83ebb4101974", - "silhouette.png": "2b74b0d3e5268524dc10a9ffa8b9d079ed0bc0e7ab7249684918470d7f2f9514", - "value.png": "f964692104759ff66053ca55cf401cbba5a6718006d02882d0679f460ceb31b8", - "detail.png": "62d19640066b41a1ac8d7a00fddeb71e31249dcc5776ef15e056ee021d368997" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.907 - }, - { - "view": "side", - "iou": 0.972 - }, - { - "view": "3/4", - "iou": 0.983 - } - ], - "focal": { - "peakPart": "body", - "peak": [ - 206, - 131 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "roof", - "share": 0.261 - }, - { - "part": "body", - "share": 0.238 - }, - { - "part": "running_board", - "share": 0.176 - } - ], - "declared": [] - }, - "rest": { - "share": 0.414, - "spread": 0.783, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.213, - "straightShare": 0.77, - "taper": -0.314, - "shape": "stable" - } + "hero.png": "dbfc00a3a2260a15cc29232e2ac64fa21a40e1adea57782672edc93ead2e40a1", + "sheet.png": "ae39d30356a96f338fb6dbea7a63d95f75ca88ef1fe35b389b0fb1e99b09e49e" } } diff --git a/recipes/brown_bear/out/brown_bear/critique.json b/recipes/brown_bear/out/brown_bear/critique.json new file mode 100644 index 00000000..0572bd05 --- /dev/null +++ b/recipes/brown_bear/out/brown_bear/critique.json @@ -0,0 +1,66 @@ +{ + "asset": "brown_bear", + "reviewer": "agent", + "overall": 5.9, + "scores": { + "silhouette": { + "score": 6, + "reason": "Reads as a big furry quadruped, but the long narrow snout, small head and long low trunk read wolverine/badger or canid more than bear in hero.png and SIDE; the shoulder hump is weak.", + "fixes": [ + "shorten and widen the muzzle and enlarge the head ~15% so the dished bear face reads", + "raise the shoulder hump mass and shorten the trunk so the back slopes from hump to rump" + ] + }, + "proportions": { + "score": 6, + "reason": "2.3 × 0.7 × 1.2 m matches the declared adult male envelope, but within it the neck is long, the head small and the hind legs stretched back like a dog's stance.", + "fixes": [ + "shorten the neck chain and bring the hind feet under the hips (plantigrade stance)", + "thicken the forearms ~15%" + ] + }, + "materials": { + "score": 6, + "reason": "Fur cards on the back and a dark-leg gradient read as fur at distance; the face and lower body are smooth painted vertex colour with one roughness.", + "fixes": [ + "extend the fur cards over the flanks and face edge so the coat is continuous", + "roughness split: glossy nose and eyes, rough coat" + ] + }, + "detail": { + "score": 6, + "reason": "Claws, nose, eyes, fur fringe on the spine; no fur clumps on the belly or legs, ear and face planes soft.", + "fixes": [ + "longer fur fringe on the belly and back of the forelegs", + "sharper brow and cheek planes; thicker, rounder ears" + ] + }, + "wear": { + "score": 4, + "reason": "Clean: no mud, no wet fur, no scars.", + "fixes": [ + "mud on the paws and lower legs (grime with the ground mask)", + "seeded scars and a rubbed patch on the shoulder" + ] + }, + "lod": { + "score": 6, + "reason": "LOD2 (1.8k) keeps the outline but is heavily faceted, loses the fur fringe and the legs become thin tubes.", + "fixes": [ + "keep a coarse fur fringe on the back at LOD2 (a few cards)", + "LOD2 legs from the sculpt with an inflate instead of tubes" + ] + }, + "technical": { + "score": 7, + "reason": "QA PASS; 160.9k / 29.4k / 1.8k triangles; 2 draws. LOD0 is heavy for a game creature and much of it is fur cards on the back.", + "fixes": [ + "cap LOD0 near 100k by coarsening the fur-card density on the back" + ] + } + }, + "renders": { + "hero.png": "6e995cad47762adf4aabdb71d3fa0dd2960dfa4c97d37eab3a102d393ae936ab", + "sheet.png": "71351a341262a42b2023e6b948ce8f85a3600492a471aa29e51326f6fc4e70ed" + } +} diff --git a/recipes/cargo_container_scifi/out/cargo_container_scifi/critique.json b/recipes/cargo_container_scifi/out/cargo_container_scifi/critique.json index 65039035..12a06f92 100644 --- a/recipes/cargo_container_scifi/out/cargo_container_scifi/critique.json +++ b/recipes/cargo_container_scifi/out/cargo_container_scifi/critique.json @@ -1,31 +1,33 @@ { "asset": "cargo_container_scifi", "reviewer": "agent", - "overall": 7.3, + "overall": 7, "scores": { "silhouette": { "score": 7, - "reason": "ISO 20 ft box with corner castings, armoured side bays, door end with four cam lock bars and status strip: reads as a container at once. The top line stays straight on purpose: anything above the top castings would stop it stacking and twist-locking (the asset's engine contract), so the critique's antenna/light bar is not added.", + "reason": "ISO 20 ft box with corner castings, armoured side bays, door end with four cam lock bars and status strip: reads as a container at once. The long side is a flat plane at grazing angles. The top line stays straight on purpose (stacking/twist-lock contract).", "fixes": [ - "deeper bay relief (8\u201310 cm) or a trapezoidal rib between bays so the long-side outline is not a flat plane at grazing angles" + "deeper bay relief (8–10 cm) or a trapezoidal rib between bays so the long-side outline is not a flat plane" ] }, "proportions": { "score": 8, - "reason": "Frame 6.058 \u00d7 2.438 \u00d7 2.591 m on ISO 668; castings 178 \u00d7 162 \u00d7 118 mm and 2 259 mm centres pass the measures; doors fill the opening between the posts.", - "fixes": [] + "reason": "Frame 6.058 × 2.438 × 2.591 m on ISO 668; castings and their centres pass the measures; doors fill the opening between the posts.", + "fixes": [ + "check the door-leaf hinge spacing against an ISO door photo (four hinges per leaf)" + ] }, "materials": { - "score": 7, - "reason": "Painted steel shell and frame (teal and off-white as paint colour, chipped to steel at the strip's chips), cast-iron corner castings, galvanised lock bars, hinges and bolts that read as metal, rubber pockets; glow on the strip and keypad. UVs now box-projected per triangle at true scale (the stretched streaks on posts are gone). Paint roughness is uniform.", + "score": 6, + "reason": "Painted steel shell and off-white frame, cast corner castings, galvanised lock bars, glow strip and keypad. But the weathering shows as large triangle-shaped grey/brown blotches on the roof and side bays in sheet.png (the vertex colour follows the triangulation), and in hero.png as soft smears rather than rust and bleaching.", "fixes": [ - "roughness variation where hands touch (door handles, keypad)", - "darker emissive housing behind the status strip" + "weather on a finer, regular vertex grid on the side and roof plates (or as a texture lane) so the colour stops following triangle shapes", + "roughness variation where hands touch (door handles, keypad)" ] }, "detail": { "score": 7, - "reason": "Bays with seams, vents, bolts, stencilled owner code on side and door, hazard bands, cam keepers and handles, hinge knuckles and pins, forklift pockets, cargo inside. Roof is ribs over a plate; no gasket line or customs seal.", + "reason": "Bays with seams, vents, bolts, stencilled owner code, hazard bands, cam keepers and handles, hinge knuckles, forklift pockets. No gasket line or customs seal; roof is ribs over a plate.", "fixes": [ "door gasket line and customs seal on the right leaf", "forklift pockets as real recesses with a lip", @@ -33,88 +35,31 @@ ] }, "wear": { - "score": 8, - "reason": "weather() (measured, in the manifest): chipped convex edges 15 %, rust streaking down from side edges 8 %, two to four dents in the side bays, bleached roof, dust on top, dirt at the sill; plus rust decals under vents and bolts and the strip's paint chips. It looks like it has crossed oceans. Doors are not weathered yet.", + "score": 7, + "reason": "Chipped edges, rust streaks from the bolts, dents, bleached roof and dirt at the sill tell a working life, but the triangle-shaped blotches undercut it and the doors are not weathered.", "fixes": [ - "run the door leaves through weather() too (hinge-space, separate call)", + "run the door leaves through weather() too (hinge space, separate call)", "faded/scraped owner code" ] }, "lod": { "score": 7, - "reason": "LOD2 (736) keeps the orange band, the four lock bars and the status strip, and the bay grid as flush quads; colours match LOD0. The roof reads lighter at LOD2 without its ribs.", + "reason": "LOD2 (736) keeps the orange band, lock bars, status strip and bay grid; the roof turns a lighter flat value split by one big diagonal shading seam.", "fixes": [ - "keep the roof ribs as quads at LOD2 so the roof value matches" + "keep the roof ribs as quads at LOD2 so the roof value matches", + "split the LOD2 roof into a regular quad grid so the diagonal seam disappears" ] }, "technical": { "score": 7, - "reason": "QA PASS incl. the weather gate; 23 022 / 8 342 / 736, one material, glow + light socket, doorArc check. LOD0 is now heavy for a prop: weather() refinement at 0.6 m splits every long bevel. Now 22 914 / 8 342 / 736. Connections gate clean (was 9 findings): rust streaks start at the panel bolts on the bay panel face, 2 mm into it, and end on that panel (they used to hover 1.5\u20135 cm off the plates and over the recesses).", + "reason": "QA PASS incl. the weather gate; 22 914 / 8 342 / 736, one material, glow + light socket, doorArc check. LOD0 is heavy for a prop because weather() refinement splits every long bevel.", "fixes": [ - "weather only the side/roof plates (not the bevelled frame) to halve the refinement cost", - "split only faces taller than one strip band in the UV pass" + "weather only the side/roof plates (not the bevelled frame) to halve the refinement cost" ] } }, "renders": { - "hero.png": "72c8f547d8a6719fec9804b355d0a7d245844ea502e12d86168dd8514c574238", - "sheet.png": "56de5ce5be8b6db63fe4fefd3b396920552a934e6344ab7ea7ff50c87af58e7d", - "raking.png": "df65b9c9a0b97f6c20822796b8464b0c8f1886183fe17cc9e74799d957103dc7", - "raking_detail.png": "b9b57c147f62594717ec8eb587aef15267c6134d67be995d3cdc7bc0a3ac3e81", - "lods.png": "0fae5692a8500258f8bc5703092f34edcc3235888a7d8f021fb0601c879262d3", - "silhouette.png": "c7ead1ff15b9b5984e77370749dc35c30510362cc7d1a489d99ba15f7a394d8a", - "value.png": "e36ea167258c8e5ff68c09f992fa3abacffa5c441613a5a0b73876afbdd71e9c", - "detail.png": "6829f58ef75b601dddcfaae00d44e12e18f19633e542823a6b258cf3960d4d5a" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 1 - }, - { - "view": "side", - "iou": 1 - }, - { - "view": "3/4", - "iou": 0.997 - } - ], - "focal": { - "peakPart": "body", - "peak": [ - 327, - 93 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "frame", - "share": 0.5 - }, - { - "part": "body", - "share": 0.449 - }, - { - "part": "casting_6", - "share": 0.037 - } - ], - "declared": [] - }, - "rest": { - "share": 0.431, - "spread": 0.816, - "min": 0.25 - }, - "shape": { - "acuteShare": 0, - "straightShare": 0.94, - "taper": -0.001, - "shape": "stable" - } + "hero.png": "cc3eb203afde4b80e29257ed56812f443e1a18db5147e50df45e338cd2bff170", + "sheet.png": "b0eeb13f4bc67b282c5b7b2aef6153bb30f70b659180afda53e9bfa853bc041d" } } diff --git a/recipes/city_block/out/city_block/critique.json b/recipes/city_block/out/city_block/critique.json new file mode 100644 index 00000000..6cb794bb --- /dev/null +++ b/recipes/city_block/out/city_block/critique.json @@ -0,0 +1,67 @@ +{ + "asset": "city_block", + "reviewer": "agent", + "overall": 5.6, + "scores": { + "silhouette": { + "score": 6, + "reason": "Reads as a dense block of brick walk-ups with flat roofs and streets from the 3/4 views. Nothing West Village-specific reaches the outline: no cornices, stoops, fire escapes, rooftop water tanks or street trees; FRONT/SIDE are a flat skyline strip.", + "fixes": [ + "projecting cornices on each building (seeded depth 0.3–0.6 m)", + "rooftop water tanks and bulkheads on the taller buildings", + "street trees along the sidewalks" + ] + }, + "proportions": { + "score": 7, + "reason": "Footprints and heights come from the pinned OSM extract (148.9 × 120 × 22.4 m), so masses are surveyed; storey heights and window sizes look right against the 10 m grid.", + "fixes": [ + "vary storey heights between buildings from their tagged levels so the window rows stop lining up across neighbours" + ] + }, + "materials": { + "score": 6, + "reason": "Brick reds, buff and brown facades and dark roofs separate by hue; windows are flat dark rectangles with no glass reflection, and the ground plane shows a fine stripe pattern.", + "fixes": [ + "a glass value on windows (glossy, slightly lighter at the top)", + "a brick strip on the facades so the walls read as brick up close", + "replace the ground stripe pattern with sidewalk slabs and asphalt" + ] + }, + "detail": { + "score": 4, + "reason": "Windows only: no doors, storefronts, stoops, fire escapes, sills/lintels or roof parapets.", + "fixes": [ + "ground-floor storefronts and doors on street-facing facades", + "stoops on the row houses and fire escapes on the walk-ups", + "window sills and lintels as shallow boxes at LOD0" + ] + }, + "wear": { + "score": 3, + "reason": "Pristine: no grime, staining, graffiti, roof patching or street dirt.", + "fixes": [ + "grime gradient at street level and soot streaks below cornices", + "patched roof membranes in varied greys" + ] + }, + "lod": { + "score": 7, + "reason": "LOD2 (2.9k) keeps every mass and its colour; windows vanish, which is acceptable at that distance.", + "fixes": [ + "bake the window grid into LOD2 vertex colour so facades do not go blank" + ] + }, + "technical": { + "score": 6, + "reason": "QA PASS; one draw; but LOD0 and LOD1 are identical (29 538 / 29 538 / 2 928): the LOD1 level saves nothing.", + "fixes": [ + "make LOD1 drop window frames/insets (ctx.lod gate) so it sits near 10k" + ] + } + }, + "renders": { + "hero.png": "d4d87b72b20c2986c65e9ff1f23650e2fac32876c062393d25b070ce2ed7bbe3", + "sheet.png": "a0aacdc3cec24a6e25fd57463d54eb698132063f169d3a17fd603dfee5d8461b" + } +} diff --git a/recipes/common_octopus/out/common_octopus/critique.json b/recipes/common_octopus/out/common_octopus/critique.json new file mode 100644 index 00000000..020cdfa2 --- /dev/null +++ b/recipes/common_octopus/out/common_octopus/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "common_octopus", + "reviewer": "agent", + "overall": 6.7, + "scores": { + "silhouette": { + "score": 8, + "reason": "Eight individually posed arms with curled tips, bulbous warty mantle and the funnel: unmistakably an octopus from every view.", + "fixes": [ + "vary the arm poses more (two arms curled under, one extended) so the TOP view is less like a symmetric star" + ] + }, + "proportions": { + "score": 7, + "reason": "1.3 × 1.4 × 0.5 m for a large adult female; arms are roughly 3× mantle length as in photos, but they stay uniformly thick for most of their length and taper late.", + "fixes": [ + "taper the arms continuously from the web to the tip (anatomy.ts)", + "enlarge the inter-arm web toward the base" + ] + }, + "materials": { + "score": 6, + "reason": "Mottled red-brown skin and pale suckers read, but the whole surface is matte: no wet sheen, no translucency at the thin arm tips.", + "fixes": [ + "lower roughness for a wet skin sheen (skin.ts)", + "pale, slightly translucent tint at the arm tips and sucker rims" + ] + }, + "detail": { + "score": 7, + "reason": "Double sucker rows along every arm, mantle warts, slit-pupil eye; the arms themselves are smooth tubes with no skin folds or papillae.", + "fixes": [ + "papillae and skin folds along the arm tops (detail.ts)", + "sucker size decreasing toward the tips" + ] + }, + "wear": { + "score": 5, + "reason": "Pristine: no missing arm tip, no scars, no colour change across the body.", + "fixes": [ + "one regenerating shorter arm tip and a few healed scars (seeded)" + ] + }, + "lod": { + "score": 6, + "reason": "LOD2 (2.1k) keeps the eight arms but the mantle turns blocky and the arm tips break into segments in the far view.", + "fixes": [ + "more rings on the LOD2 mantle and arm tips (~3k total)" + ] + }, + "technical": { + "score": 8, + "reason": "QA PASS; 133.7k / 13.4k / 2.1k triangles; 2 draws.", + "fixes": [ + "move LOD0 triangles from sucker rims on the hidden underside to the mantle" + ] + } + }, + "renders": { + "hero.png": "9f66d0eae35da43d211d07b1057d256a39048977ca846be06e7293a03945a34b", + "sheet.png": "777fb13f95be373a7db979ac4a0d8f836d64c1fc33b366f70f907b901e99bb56" + } +} diff --git a/recipes/dragon/out/dragon/critique.json b/recipes/dragon/out/dragon/critique.json new file mode 100644 index 00000000..0b21cf50 --- /dev/null +++ b/recipes/dragon/out/dragon/critique.json @@ -0,0 +1,67 @@ +{ + "asset": "dragon", + "reviewer": "agent", + "overall": 6, + "scores": { + "silhouette": { + "score": 7, + "reason": "Bat wings, ringed swept-back horns, dorsal spikes and spade tail: reads as a dragon at once. The body is lumpy and potato-like and the legs short and stubby, so the outline reads cartoon rather than powerful.", + "fixes": [ + "smooth the muscle noise on the trunk (body.ts) so the outline follows big masses, not lumps", + "lengthen the legs ~15% and narrow the lower legs (legs.ts)" + ] + }, + "proportions": { + "score": 6, + "reason": "8.7 × 7.9 × 4.7 m: wings and head are in proportion, but the neck is short, the legs stubby and the tail thin compared with the body mass.", + "fixes": [ + "lengthen the neck chain by one vertebra (neck.ts)", + "thicken the tail root to match the hips (tail.ts)" + ] + }, + "materials": { + "score": 6, + "reason": "Crimson vertex colour with cream belly plates and darker wing membrane; the surface noise reads as clay lumps rather than scales, all at one roughness; horns and teeth separate by hue only.", + "fixes": [ + "a scale pattern on the back and flanks (scales.ts stamps at LOD0)", + "roughness split: glossy eyes and teeth, rough membrane" + ] + }, + "detail": { + "score": 6, + "reason": "Teeth, horns with rings, dorsal spikes, wing fingers, claws; no individual scales, membrane veins only painted, belly plates soft.", + "fixes": [ + "stamped scale rows on the neck and back", + "raised veins on the wing membranes", + "crisper belly-plate edges" + ] + }, + "wear": { + "score": 4, + "reason": "Pristine: no scars, no torn membrane, no chipped horns or soot.", + "fixes": [ + "a torn notch in one wing membrane and scars on the flank", + "chipped horn tips and soot around the jaw" + ] + }, + "lod": { + "score": 6, + "reason": "LOD2 (4k) keeps the pose but the wings become flat triangles, the tail a thin stick and the body faceted.", + "fixes": [ + "keep the wing finger bones at LOD2 as thin quads so the wing outline holds", + "more rings on the LOD2 tail" + ] + }, + "technical": { + "score": 7, + "reason": "QA PASS; 198.6k / 44.2k / 4.1k triangles, 6 draws. LOD0 is heavy for a game creature.", + "fixes": [ + "cap LOD0 near 120k by coarsening the hidden belly and inner-leg sculpt cells" + ] + } + }, + "renders": { + "hero.png": "6f53470be1a30b7d4410af268f477ed6e3166589822af2b6b45ca896a2366636", + "sheet.png": "b65d37486ed01258442fd51763c4cbcfa97da3c94fd964fdcd343ce833ca49ac" + } +} diff --git a/recipes/dragon_hybrid/README.md b/recipes/dragon_hybrid/README.md index 2f79737f..cdc323c8 100644 --- a/recipes/dragon_hybrid/README.md +++ b/recipes/dragon_hybrid/README.md @@ -46,6 +46,12 @@ Measured on the build: shoulder 1.71 m, head and body 3.24 m, tail bones 2.20 m. Five cached requests, each naming the previous output by sha256 (kinds `modelChain`). A build never calls a model or Blender; `t3d neural run` does. +The chain is `required: true`. The code quadruped this spec would give without the model is not this +asset (over 100 000 triangles at LOD 0 against a 30 000 budget, off the sourced box), so `t3d build` +with an output missing stops with one `T3D_NEURAL_MISSING` and builds nothing. The npm package does +not ship `neural/`: clone the repository to build this recipe, or copy the folder into your project +and run `t3d neural run` on the copy (GPU and Blender, below). + | # | Task | Tool (licence) | Output | In git | |---|---|---|---|---| | 1 | `image` | Z-Image-Turbo (Apache-2.0), seed 9, 1024² | concept PNG, 0.66 MB | yes | diff --git a/recipes/dragon_hybrid/dragon_hybrid.test.ts b/recipes/dragon_hybrid/dragon_hybrid.test.ts index bc795e0b..d2750679 100644 --- a/recipes/dragon_hybrid/dragon_hybrid.test.ts +++ b/recipes/dragon_hybrid/dragon_hybrid.test.ts @@ -7,7 +7,14 @@ import { afterEach, describe, expect, it } from 'bun:test'; import { join } from 'node:path'; -import { clearNeuralSources, compile, runQa, toGlb } from '@developerz.ai/ts23d'; +import { + clearNeuralSources, + compile, + neuralCache, + provideNeural, + runQa, + toGlb, +} from '@developerz.ai/ts23d'; import { compileLods } from '@developerz.ai/ts23d/testing'; import { provideRecipeNeural } from '../../src/cli/neural-store.ts'; import { dragon_hybrid, measured, SOURCED } from './dragon_hybrid.ts'; @@ -40,6 +47,12 @@ describe('dragon_hybrid, CPU tier (no neural cache)', () => { }); describe('dragon_hybrid, model tier (committed cache)', () => { + it('a build with an empty cache (the npm package) stops with T3D_NEURAL_MISSING, no CPU stand-in', () => { + clearNeuralSources(); + provideNeural(neuralCache({ recipe: 'dragon_hybrid', offline: true })); + expect(() => compile(dragon_hybrid, { lod: 2 })).toThrow('T3D_NEURAL_MISSING'); + }); + it( 'a missing cache is not mistaken for the model: the textured body appears only with it', async () => { diff --git a/recipes/dragon_hybrid/out/dragon_hybrid/critique.json b/recipes/dragon_hybrid/out/dragon_hybrid/critique.json new file mode 100644 index 00000000..5f02eb3c --- /dev/null +++ b/recipes/dragon_hybrid/out/dragon_hybrid/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "dragon_hybrid", + "reviewer": "agent", + "overall": 7.1, + "scores": { + "silhouette": { + "score": 8, + "reason": "Raised bat wings with finger bones, horned head, long neck and spiked tail on four legs: reads as a dragon from every view.", + "fixes": [ + "longer wing fingers so the wings read clearly as a flying surface in FRONT" + ] + }, + "proportions": { + "score": 7, + "reason": "4.9 × 2.8 × 3.3 m with the tail; the body and neck sit well, but the wings are small for a 600 kg flyer and the legs a little thin for the mass.", + "fixes": [ + "scale the wings ~25% in a Blender step so the span is plausible for the body", + "thicken the legs ~10%" + ] + }, + "materials": { + "score": 7, + "reason": "Image-to-3D texture reads as scales with cream belly plates and a lighter wing membrane; the horns are flat chalky grey and lighting is baked in.", + "fixes": [ + "repaint the horns with a keratin gradient (dark base to pale tip)", + "de-light the albedo before import" + ] + }, + "detail": { + "score": 7, + "reason": "Scale texture, belly plates, claws and wing bones hold at third-person distance; the horns are blobby and the claws painted rather than modelled.", + "fixes": [ + "replace the horns with code-built ringed cones on the head bone", + "model claws on each foot" + ] + }, + "wear": { + "score": 6, + "reason": "Some colour variation; no torn membrane, no scars or chipped horns.", + "fixes": [ + "a torn notch in one wing membrane and scars on the flank (seeded)" + ] + }, + "lod": { + "score": 8, + "reason": "LOD2 (5k) is almost identical to LOD0 at the far view.", + "fixes": [ + "drop LOD2 to ~3k: the far view does not need 5k" + ] + }, + "technical": { + "score": 7, + "reason": "QA PASS; 30k / 10k / 5k triangles, one draw per LOD; idle/look/walk clips, but the wings ride with the chest and never flap.", + "fixes": [ + "add a wing bone chain and a flap clip so the wings move" + ] + } + }, + "renders": { + "hero.png": "3d9df4a7dbb6765e6a87c479258f9aac1a2f3bce0077b32dae31df879cc66397", + "sheet.png": "0877376dd51f1bef9de5a117762aadfcfee8f063babf8d43558acda28880f27f" + } +} diff --git a/recipes/dragon_hybrid/surface.ts b/recipes/dragon_hybrid/surface.ts index 8d02d0d1..b58e79f0 100644 --- a/recipes/dragon_hybrid/surface.ts +++ b/recipes/dragon_hybrid/surface.ts @@ -9,7 +9,10 @@ // elbows and the spine lie (mapped by topology) // blender FINAL: transform(fit) + steps.ts → the game mesh: three LODs, PNG maps, the skin; // committed, the only thing a build draws -// A miss anywhere returns null: the code dragon (dragon_hybrid.ts). Nothing here calls a model; +// The chain is `required`: the code creature of this spec fails QA (over 100 000 triangles at +// LOD 0, off the sourced box), so `t3d build` with an output missing stops with one +// T3D_NEURAL_MISSING and never ships it. A compile with no cache provided at all (a test, the +// gallery's CPU tier) still gets null: the code dragon (dragon_hybrid.ts). Nothing here calls a model; // `t3d neural run` does, on a machine with a GPU and Blender (README.md). import { @@ -39,7 +42,9 @@ export const SURFACE: Omit = { * QA pass on this skin (its wrist folds 146°); trot and gallop go further. */ export const DROPPED: readonly string[] = ['trot', 'gallop']; -/** The imported game mesh with the rig it was built for, or null on any cache miss. */ +/** The imported game mesh with the rig it was built for; null with no cache provided + * and while `t3d neural run` is still filling the cache; an offline (build) cache throws + * `T3D_NEURAL_MISSING` on a miss instead. */ export const dragonModel = (ctx: BuildContext): ChainedModel | null => modelChain(ctx, { concept: CONCEPT, @@ -55,6 +60,7 @@ export const dragonModel = (ctx: BuildContext): ChainedModel | null => maxTriangles: TRIANGLES[0], models: { image: CONCEPT_MODEL, mesh: SURFACE_MODEL }, license: 'MIT', + required: true, author: 'ts23d dragon_hybrid recipe (Z-Image concept, Pixal3D mesh, UniRig joints, Blender steps)', }); diff --git a/recipes/dropship/out/dropship/critique.json b/recipes/dropship/out/dropship/critique.json index 0139c698..4be2e704 100644 --- a/recipes/dropship/out/dropship/critique.json +++ b/recipes/dropship/out/dropship/critique.json @@ -1,29 +1,34 @@ { "asset": "dropship", "reviewer": "agent", - "overall": 7.3, + "overall": 7.1, "scores": { "silhouette": { "score": 8, - "reason": "Original VTOL dropship: lofted nose with framed canopy, slab cabin with rear ramp, two tilting ducted nacelles on sponsons, swept tail and fin, landing gear, chin gun. Distinct and readable from every view and at LOD2.", - "fixes": [] + "reason": "Original VTOL dropship: lofted nose with framed canopy, slab cabin with rear ramp, two tilting ducted nacelles on sponsons, swept tail and fin, landing gear. Distinct and readable from every view and at LOD2.", + "fixes": [ + "a deeper chin or sensor blister under the nose so the FRONT view is less of a plain oval" + ] }, "proportions": { "score": 8, "reason": "Design dossier scale (cabin fits troops at 1.8 m, ramp at walking slope) and envelope pass; nacelles in proportion to the hull.", - "fixes": [] + "fixes": [ + "shorten the tail boom ~0.5 m so the nacelles sit nearer the centre of mass in SIDE" + ] }, "materials": { - "score": 7, - "reason": "Hull plates now steel_painted (chips to bare metal), aluminium mechanisms, cast fittings, glass canopy, glowing thrusters and navigation lights, hazard and accent paint. The chip pattern is uniform and too dense on the large plates; the nose loft shows triangle faceting in the shading.", + "score": 6, + "reason": "Steel-painted hull with chips, aluminium mechanisms, glass canopy, glowing lights, hazard and accent paint. In hero.png the left canopy pane shows a patch of white speckle noise, the nose loft shows triangle faceting in the shading, and the chip pattern is uniform and dense on the large plates.", "fixes": [ - "lower chip density on big plates (a defined steel_painted variant)", - "weighted normals on the nose loft" + "find and remove the white speckle on the canopy pane (glass coplanar with the frame/interior, or grime on glass)", + "weighted normals on the nose loft", + "lower chip density on big plates (a defined steel_painted variant)" ] }, "detail": { "score": 7, - "reason": "Plate seams, vents and louvres, greebles, antennae, hatches, stencils and tail number, gear with pads, nacelle hazard rings and louvres, cabin interior and lamp behind the ramp.", + "reason": "Plate seams, vents and louvres, greebles, antennae, hatches, stencils and tail number, gear with pads, nacelle hazard rings and louvres, cabin interior.", "fixes": [ "rivet/fastener rows along the plate seams", "hydraulic lines on the gear legs" @@ -31,9 +36,9 @@ }, "wear": { "score": 6, - "reason": "Chipped paint everywhere, landing-pad dirt on gear and belly, dust on top. No exhaust scorch around the nacelles, no oil streaks, no faded stencils.", + "reason": "Chipped paint everywhere, dirt on gear and belly, dust on top. No exhaust scorch around the nacelles, no oil streaks, no faded stencils.", "fixes": [ - "heat scorch inside and below the nacelle mouths (soot zones)", + "heat scorch inside and below the nacelle mouths", "oil streaks from the gear bays" ] }, @@ -46,71 +51,14 @@ }, "technical": { "score": 8, - "reason": "QA PASS, 18 810 / 6 460 / 2 000, glow rules, ramp arc check, moving nacelles and ramp clear. The UV split for true-scale textures cost ~4k LOD0 triangles. Now 18 970 / 6 516 / 2 032. Connections gate clean (was 30 findings): troop-seat backrests sit in their pans and against the wall panels; the nacelle bore runs on to key the fan's stator disc; exhaust vanes span the nozzle bore chord to chord.", + "reason": "QA PASS, 18 766 / 6 458 / 2 002; glow rules, ramp arc check, moving nacelles and ramp clear; connections gate clean.", "fixes": [ - "split only faces taller than one strip band" + "split only faces taller than one strip band in the UV pass" ] } }, "renders": { - "hero.png": "3bb66d286761cefd38ad4efb1a5b2539c7b1977d37e174c6291e26c57efec0d2", - "sheet.png": "54fcf281d1c4cc15afb2c0822a43396155d06a041d6a231ddf3b67e17d803f54", - "raking.png": "18aaf7373ef2bd85de23a7566ff61eabd0b1b4032e55aa6fc76c21e889f76532", - "raking_detail.png": "bbba2f69e244aad48a5e7b2ebcae67b650f330bf01902a87406f2142657e56c5", - "lods.png": "c48e939f228d876a5fc3a6e782c2bc92cb50f66f24f2290b71d0d2f12ca9362e", - "silhouette.png": "92605fe555be2802ea6c330e6e85199d48a0a7e7649a4b3ae2f36798bd5b2f14", - "value.png": "ded60cbfb28dce6e1060fabdb2a2ac5c1d296eae038d1a9d0b55eaeef4721556", - "detail.png": "44103d5c4d444df75c440d697add34ed8fbad32c40de29ccacab940476545b8b" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.985 - }, - { - "view": "side", - "iou": 0.984 - }, - { - "view": "3/4", - "iou": 0.987 - } - ], - "focal": { - "peakPart": "canopy", - "peak": [ - 537, - 261 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "canopy", - "share": 0.406 - }, - { - "part": "dorsal", - "share": 0.195 - }, - { - "part": "hull_detail", - "share": 0.167 - } - ], - "declared": [] - }, - "rest": { - "share": 0.624, - "spread": 0.704, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.475, - "straightShare": 0.572, - "taper": -0.195, - "shape": "aggressive" - } + "hero.png": "7340ecfaecdba8fcd4422c5e8565762ef7ed49d6e6da640554ea621c96f288a0", + "sheet.png": "2ffd71cca1826548af99d6bc5e944de38a144f38c06054bde27d7459266f1efa" } } diff --git a/recipes/eiffel_tower/out/eiffel_tower/critique.json b/recipes/eiffel_tower/out/eiffel_tower/critique.json index 80d67a31..10ae6db1 100644 --- a/recipes/eiffel_tower/out/eiffel_tower/critique.json +++ b/recipes/eiffel_tower/out/eiffel_tower/critique.json @@ -6,30 +6,34 @@ "silhouette": { "score": 9, "reason": "Unmistakable: four lattice legs on masonry piers, the first-floor arches, three decks, the tapering upper shaft, campanile lantern and antenna. Reads in pure black.", - "fixes": [] + "fixes": [ + "open the lattice between the upper-shaft corner posts a little more so the shaft reads see-through against the sky as in photos" + ] }, "proportions": { "score": 8, - "reason": "Height to the antenna, base span and deck heights pass the sourced measures (fr.wikipedia, Eiffel's plates).", - "fixes": [] + "reason": "Height to the antenna, base span and deck heights pass the sourced measures (124.9 × 125 × 330.6 m).", + "fixes": [ + "check the second-deck height and width against the elevation plate by overlay" + ] }, "materials": { "score": 7, - "reason": "Lattice now painted wrought iron (steel_painted under the three-tone Eiffel Tower Brown gradient by height, chips showing rusty iron), stone piers, glazed pavilions. The paint film reads uniform at hero distance.", + "reason": "Painted wrought iron under the three-tone brown gradient, stone piers, glazed pavilions; the paint film reads uniform at hero distance.", "fixes": [ "slightly glossier, fresher paint on the upper shaft (repainted most often)" ] }, "detail": { "score": 8, - "reason": "188k LOD0 triangles: lattice girders with zig-zag webs, arches, decks with railings, pavilions, lift housings, campanile ribs and side arches (base ring now wide enough to carry the rib feet).", + "reason": "188k LOD0 triangles: lattice girders with zig-zag webs, arches, decks with railings, pavilions, lift housings, campanile ribs.", "fixes": [ - "mid rails on the deck railings with square posts (dropped: diamond posts only met them along an edge)" + "mid rails on the deck railings with square posts" ] }, "wear": { "score": 6, - "reason": "Grime at the feet, paint chips from the strip; no rust bleeding at rivet lines or pigeon streaks on the decks.", + "reason": "Grime at the feet, paint chips; no rust bleeding at rivet lines or streaks on the decks.", "fixes": [ "weather() rust streaks below the lattice nodes on the lower legs", "dirt on the deck tops" @@ -37,74 +41,21 @@ }, "lod": { "score": 7, - "reason": "LOD2 (1 062) keeps the silhouette shell and decks; the lattice becomes solid frustums.", - "fixes": [] + "reason": "LOD2 (1 052) keeps the silhouette shell and decks, but the lattice becomes solid brown frustums: the far view reads as a solid sculpture, not a see-through tower.", + "fixes": [ + "an alpha-cutout lattice strip on the LOD2 frustums so it stays see-through at distance" + ] }, "technical": { "score": 8, - "reason": "QA PASS incl. the new connections gate (campanile ribs now stand on their ring; railing mid rails removed): 187 872 / 21 860 / 1 062, one draw.", - "fixes": [] + "reason": "QA PASS incl. connections; 187 768 / 21 796 / 1 052, one draw.", + "fixes": [ + "instance the repeated lattice panels (EXT_mesh_gpu_instancing) to cut LOD0 file size" + ] } }, "renders": { - "hero.png": "1f8175fae00248e1e5807e26abab217872847627def92317efd86d39ba7e8593", - "sheet.png": "90526542b5dfd0f5543632478f5844b2a1455f0d481ac021385ef11223c5a642", - "raking.png": "4adf7a39d497759c939295a1d5a92c2e98ff96c48066fc1d6fe21abfb8afa5df", - "raking_detail.png": "7b41ec742ef223d59b12370ed3cbca854ab077f9e51a4f36a1138b46909ef341", - "lods.png": "5ceb1b6088b89b3e6f37a0840b2dbbe3e80dc1067816a12675d35a6236f1674f", - "silhouette.png": "835054d2d96de0c194eac8501fda8e02417cebf1dd06b429edb5b6086441414c", - "value.png": "6257d7458d87e8abf8bd3adc280370fd3f9e3f36ce4b10a7d153a4e0dce312ef", - "detail.png": "dfaeadf7e429e772cef3756a80d4c4d9225086ba76b73667903de31921a37a92" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.44 - }, - { - "view": "side", - "iou": 0.379 - }, - { - "view": "3/4", - "iou": 0.845 - } - ], - "focal": { - "peakPart": "campanile", - "peak": [ - 398, - 94 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "leg_2", - "share": 0.195 - }, - { - "part": "pavilion_1", - "share": 0.182 - }, - { - "part": "leg_3", - "share": 0.139 - } - ], - "declared": [] - }, - "rest": { - "share": 0, - "spread": 0.041, - "min": 0.25 - }, - "shape": { - "acuteShare": 1, - "straightShare": 1, - "taper": 0.851, - "shape": "aggressive" - } + "hero.png": "fb28196fd72b6c84719c9d07a23eb94ba22cfb55a9d7f0b8aebaead4bf6e9ff7", + "sheet.png": "9e5577e97ab0cdc3b535c01fad0cf49e5d5456184f981f0a6498d3a513d628df" } } diff --git a/recipes/elf_head/out/elf_head/critique.json b/recipes/elf_head/out/elf_head/critique.json index a864bebc..56df3aad 100644 --- a/recipes/elf_head/out/elf_head/critique.json +++ b/recipes/elf_head/out/elf_head/critique.json @@ -1,34 +1,35 @@ { "asset": "elf_head", "reviewer": "agent", - "overall": 5.9, + "overall": 5.4, "scores": { "silhouette": { "score": 7, - "reason": "Pointed ears and long locks name an elf from any side; LOD2 IoU high.", + "reason": "Pointed ears and long fair hair over a cloaked bust read as an elf in FRONT and SIDE. In hero.png the hair is sparse strands splaying outward from the head like wet string, and the ears hide behind it at 3/4.", "fixes": [ - "more hair volume at the back of the head (the cap reads as a helmet)" + "more hair volume over the crown and denser locks (hair.ts) so the hair reads as a mass, not strands", + "part the hair behind the ears so the ear tips break the outline at 3/4" ] }, "proportions": { "score": 6, - "reason": "Jaw no longer double-narrowed, full cheeks, lips back in a straight profile, ribbons clear of the face; the face is still long and the brow/orbit area reads older than intended.", + "reason": "Full cheeks, straight profile, slim neck; the face is still long and the orbit area reads older than intended.", "fixes": [ - "shorten the mid-face 3 %", + "shorten the mid-face 3% (spec.ts)", "raise the lid crease and thin the upper lid" ] }, "materials": { "score": 5, - "reason": "Hair, skin, tunic, cloak and trim separate by hue; no roughness split.", + "reason": "Hair, skin, tunic, cloak and trim separate by hue, with no roughness split; the hair strands read as thin plastic; the tunic trim stair-steps at the vertex cell in FRONT/SIDE.", "fixes": [ - "silk/wool strips for tunic and cloak", - "metal finish on the clasp" + "silk/wool strips for tunic and cloak and a metal finish on the clasp", + "trim colour on a finer cell (or as geometry) so it stops stair-stepping" ] }, "detail": { "score": 6, - "reason": "Ear structure, lids, ribbon locks with strand grooves, tall blended collar, hemmed cloak; hair cap lacks strand detail.", + "reason": "Ear structure, lids, strand locks, tall collar, hemmed cloak and trim, clasp; no embroidery, no strand detail over the crown.", "fixes": [ "strand grooves over the crown", "embroidery on the trim" @@ -38,93 +39,27 @@ "score": 3, "reason": "Pristine.", "fixes": [ - "dust on the cloak shoulders", - "fray at the hems" + "dust on the cloak shoulders and fray at the hems" ] }, "lod": { - "score": 7, - "reason": "Outline and blocks hold at LOD2.", + "score": 4, + "reason": "LOD2 in sheet.png is a different character: the long blond strands are replaced by a short brown cap, the face turns blotchy and the cloak trim breaks up. The switch would visibly pop.", "fixes": [ - "keep the ribbons at LOD2 as a flat strip" + "keep the long hair at LOD2 as a few flat blond cards (hair.ts lod === 2 path) in the LOD0 colour", + "carry the LOD0 hair and trim colours to LOD2 instead of the brown base cap" ] }, "technical": { "score": 7, - "reason": "QA PASS honestly with 987 small crease regions (≤ 2.5 cm, from the geometry); socket joints pass; 79k / 24k / 3.8k tris.", + "reason": "QA PASS; 100.1k / 23.4k / 3.4k triangles; 4 / 4 / 3 draws.", "fixes": [ - "merge duplicate head/bust crease spheres" + "merge duplicate head/bust crease spheres and drop hidden scalp under the hair at LOD0" ] } }, "renders": { - "hero.png": "95eb8e90c2adcc9a81e30f2f574ea0584d6267520d33a0444098cca2849cfd32", - "sheet.png": "5111bbacda8598350a63ed57fb44d2019d9a4db426a8e3aa095143785b6c29fb", - "raking.png": "521ecbae8f3d9e4f6f79915c094ee03bd7944b754476899b516f4d50d91301ca", - "raking_detail.png": "1f82a6e63e660c179f84bd0e6f6073e49d5fd855ccd8706f2595d39833aa58a2", - "lods.png": "943af0590ad689e4cf4ae736c5cf0f4786c9ae98a5e2587ddc92a18640f29172", - "silhouette.png": "0fedab9f8aedc86670b9a73b9b4186dd77d3e03ea4dc7de1aec9947aeac1404b", - "value.png": "3a50082dc881ac7369eb6f082310ee57c7a2d790f398fe8493abec388c7997de", - "detail.png": "295fec6b59167a254eff9dff4f5e052de64ddc2793c032eca2899cf7059b70cc" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.93 - }, - { - "view": "side", - "iou": 0.946 - }, - { - "view": "3/4", - "iou": 0.92 - } - ], - "focal": { - "peakPart": "head", - "peak": [ - 424, - 126 - ], - "focalShare": 0.987, - "focalArea": 0.529, - "top": [ - { - "part": "head", - "share": 0.987 - }, - { - "part": "eye_r", - "share": 0.013 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.434, - "spread": 0.608, - "min": 0.25 - }, - "grounded": { - "rise": 0.39, - "bands": [ - 0.154, - 0.274, - 0.38, - 0.453, - 0.455 - ] - }, - "shape": { - "acuteShare": 0.126, - "straightShare": 0.459, - "taper": 0.151, - "shape": "friendly", - "declared": "friendly" - } + "hero.png": "62462d89dcb42c5bc840afbb90508ec3c5f7c31e0f6da0032c0da0d14c791c01", + "sheet.png": "06725f5fcef779107bd515759b9f332699e5640295e028c84d8055d734dcc452" } } diff --git a/recipes/face_demo/out/face_dwarf/critique.json b/recipes/face_demo/out/face_dwarf/critique.json index 6194183e..52415dea 100644 --- a/recipes/face_demo/out/face_dwarf/critique.json +++ b/recipes/face_demo/out/face_dwarf/critique.json @@ -1,134 +1,64 @@ { "asset": "face_dwarf", "reviewer": "agent", - "overall": 5, + "overall": 5.1, "scores": { "silhouette": { "score": 6, - "reason": "Spade beard, broad nose, wide jaw and thick neck read as a dwarf; the beard body is a good mass but its sides are slab-like and the locks are tube-like.", + "reason": "Spade beard, broad nose, wide jaw and thick neck read as a dwarf; the beard sides are slab-like and the locks tube-like; the hero reads closer to a ginger man than a dwarf.", "fixes": [ - "shape the beard sides with a taper and let locks overlap the mass edge", - "braid two front locks" + "taper the beard sides and let locks overlap the mass edge; braid two front locks" ] }, "proportions": { "score": 6, - "reason": "Short face (×0.88), broad nose (×1.4), wide jaw (×1.25) hold; eye spacing 64.5 vs 65.3 mm. Older age (0.6) shows only in nose and lobes.", + "reason": "Short face, broad nose and wide jaw hold; eye spacing on spec. Older age shows only in nose and lobes.", "fixes": [ "stronger age: deeper nasolabial folds and lid sag as displacement" ] }, "materials": { "score": 4, - "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets, curvature cavity, mottle) separate by hue but not by roughness or SSS; hair is the same material as skin, so it reads as sculpted clay; eyes are the only glossy part (cornea clearcoat).", + "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets) separate by hue but not by roughness or subsurface; the hair is the same material as the skin, so it reads as sculpted clay; there is a pale band of bare skin under the hair helmet edge at the hairline in hero.png.", "fixes": [ - "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen (material-artist)", - "darker hair roots and per-lock value variation along the lock (root → tip colour is set but too subtle)" + "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen", + "darker hair roots and per-lock value variation along each lock", + "close the hairline: extend the hair mass to the skin (or tint the scalp under it to the root colour) so no pale band shows" ] }, "detail": { "score": 5, - "reason": "Primary and secondary forms come from a real anatomical base (eyelids, nasolabial area, lips, ear helix/antihelix/concha, clavicles, sternocleidomastoid) and read cleanly under the raking light, but there is no tertiary detail: no pores, wrinkles, lid creases or brow hairs (brows are painted bands), and the SDF hair locks are uniform grooves.", + "reason": "Primary and secondary forms come from a real anatomical base (lids, nasolabial area, lips, ear helix, clavicles, sternocleidomastoid), but there is no tertiary layer (pores, wrinkles) and the hair is evenly spaced grooves.", "fixes": [ - "LOD0 normal-scale wrinkle/pore displacement on the subdivided base driven by age", - "hair: varied lock widths and a parting instead of evenly spaced grooves; hair cards (planned)" + "LOD0 wrinkle/pore displacement on the subdivided base driven by age", + "varied lock widths and a parting instead of evenly spaced grooves" ] }, "wear": { "score": 3, "reason": "Pristine: seeded asymmetry and age/weight knobs are the only story; no scars, stubble, blemishes, sun-darkened skin or dirt.", "fixes": [ - "seeded scars/blemish decals in vertex colour (orc, dwarf)", - "stubble shadow on the human/dwarf jaw", - "age-driven darkening under the eyes" + "seeded scar/blemish marks in vertex colour", + "stubble shadow on the jaw and age-driven darkening under the eyes" ] }, "lod": { - "score": 6, - "reason": "LOD2 keeps the beard and hair blocks; face soft.", + "score": 5, + "reason": "LOD2 keeps the beard and hair blocks, but the face breaks into blotchy patches around the eyes and the bottom edge of the bust turns ragged in sheet.png.", "fixes": [ - "authored LOD2 topology" + "LOD2 from an authored coarse topology instead of the distance-field re-mesh (it cuts the face colour into patches)" ] }, "technical": { "score": 7, - "reason": "QA PASS at every LOD; 145k/61k/11.5k tris (beard + hair are ~65 %); 2 draws.", + "reason": "QA PASS at every LOD; 144.4k / 60.9k / 11.4k tris (beard + hair are most of it); 2 draws.", "fixes": [ "coarser beard sculpt cell at LOD1" ] } }, "renders": { - "hero.png": "8f793b82f1a433a84fc57b784b92bceac1b7d18fb52d970868688bca02435467", - "sheet.png": "1e43fd7370fdefb44f342b5ed61cbb2e3c69e96a61446af1d5266e68275fbdd1", - "raking.png": "cc3e9c281b512b200e918d01d513eb8e826f0249edfa382214e665b32746ed0f", - "raking_detail.png": "1950231f462c47ac3ce0685cc550367d44f20652a53f61f7f171ce941d79fce6", - "lods.png": "919fce659d7a07b59fef5a55a4d3d456741b9e41ff9b8a9c2b23316c75f7a7ce", - "silhouette.png": "85e1410292593bb43f3036bdc80692bef84a9d61cc5ee5af2cac6e0a765d2d5d", - "value.png": "dace1917d3e9e86fd3f05088a27a5a823f198a6550f16859b37088ce493199dd", - "detail.png": "a19ac01b6a5e3994054df6404469d90de0154e3b9245fe3aed6eccb2bd2dcd09" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.979 - }, - { - "view": "side", - "iou": 0.954 - }, - { - "view": "3/4", - "iou": 0.961 - } - ], - "focal": { - "peakPart": "skin", - "peak": [ - 466, - 181 - ], - "focalShare": 0.687, - "focalArea": 0.569, - "top": [ - { - "part": "skin", - "share": 0.687 - }, - { - "part": "plinth", - "share": 0.162 - }, - { - "part": "hair", - "share": 0.15 - } - ], - "declared": [ - "skin" - ] - }, - "rest": { - "share": 0.589, - "spread": 0.655, - "min": 0.25 - }, - "grounded": { - "rise": 0.093, - "bands": [ - 0.343, - 0.54, - 0.513, - 0.478, - 0.467 - ] - }, - "shape": { - "acuteShare": 0.1, - "straightShare": 0.473, - "taper": 0.23, - "shape": "friendly" - } + "hero.png": "02b691e3f5a66569815fadfeebf6dbf894bf4291b22e2ef33f6257c77f3565a5", + "sheet.png": "2b532e4f04b2341ea803a10ae2a4a121371012640b334b9879b6a9501e8dd744" } } diff --git a/recipes/face_demo/out/face_elf/critique.json b/recipes/face_demo/out/face_elf/critique.json index 2bd62f6c..ea5e1911 100644 --- a/recipes/face_demo/out/face_elf/critique.json +++ b/recipes/face_demo/out/face_elf/critique.json @@ -1,11 +1,11 @@ { "asset": "face_elf", "reviewer": "agent", - "overall": 5, + "overall": 5.1, "scores": { "silhouette": { "score": 6, - "reason": "Long refined face and pointed back-swept ears read as an elf; the long hair falls behind the ears in ribbons. The hair cap is too tight and helmet-like for long hair.", + "reason": "Long refined face and pointed ears read as an elf; the hair is a tight grooved helmet with two flat ribbons behind the ears.", "fixes": [ "fuller hair volume over the crown and ribbons that fan over the shoulders", "lengthen the ear tip another 1–2 cm" @@ -13,122 +13,53 @@ }, "proportions": { "score": 7, - "reason": "Jaw ×0.85, face ×1.08, high zygoma and refined nose read as declared; eye spacing 63.7 vs 64.1 mm.", + "reason": "Narrow jaw, long face, high zygoma and refined nose read as declared; eye spacing on spec.", "fixes": [ - "stronger almond tilt of the eyes (tilt 0.03 H)" + "stronger almond tilt of the eyes" ] }, "materials": { "score": 4, - "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets, curvature cavity, mottle) separate by hue but not by roughness or SSS; hair is the same material as skin, so it reads as sculpted clay; eyes are the only glossy part (cornea clearcoat).", + "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets) separate by hue but not by roughness or subsurface; the hair is the same material as the skin, so it reads as sculpted clay; there is a pale band of bare skin under the hair helmet edge at the hairline in hero.png.", "fixes": [ - "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen (material-artist)", - "darker hair roots and per-lock value variation along the lock (root → tip colour is set but too subtle)" + "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen", + "darker hair roots and per-lock value variation along each lock", + "close the hairline: extend the hair mass to the skin (or tint the scalp under it to the root colour) so no pale band shows" ] }, "detail": { "score": 5, - "reason": "Primary and secondary forms come from a real anatomical base (eyelids, nasolabial area, lips, ear helix/antihelix/concha, clavicles, sternocleidomastoid) and read cleanly under the raking light, but there is no tertiary detail: no pores, wrinkles, lid creases or brow hairs (brows are painted bands), and the SDF hair locks are uniform grooves.", + "reason": "Primary and secondary forms come from a real anatomical base (lids, nasolabial area, lips, ear helix, clavicles, sternocleidomastoid), but there is no tertiary layer (pores, wrinkles) and the hair is evenly spaced grooves.", "fixes": [ - "LOD0 normal-scale wrinkle/pore displacement on the subdivided base driven by age", - "hair: varied lock widths and a parting instead of evenly spaced grooves; hair cards (planned)" + "LOD0 wrinkle/pore displacement on the subdivided base driven by age", + "varied lock widths and a parting instead of evenly spaced grooves" ] }, "wear": { "score": 3, "reason": "Pristine: seeded asymmetry and age/weight knobs are the only story; no scars, stubble, blemishes, sun-darkened skin or dirt.", "fixes": [ - "seeded scars/blemish decals in vertex colour (orc, dwarf)", - "stubble shadow on the human/dwarf jaw", - "age-driven darkening under the eyes" + "seeded scar/blemish marks in vertex colour", + "stubble shadow on the jaw and age-driven darkening under the eyes" ] }, "lod": { - "score": 5, - "reason": "LOD2 keeps the outline but the thin ear tips and ribbons erode at the 1.4 cm cell.", + "score": 4, + "reason": "LOD2 in sheet.png loses the ear tips and the hair ribbons entirely, the mouth area turns into dark patches and the bust edge is ragged: the elf no longer reads as an elf at distance.", "fixes": [ - "keep ears from the LOD1 mesh at LOD2 (authored topology) or thicken thin parts before the re-mesh" + "keep the ears and ribbons from the LOD1 mesh at LOD2, or thicken thin parts before the re-mesh" ] }, "technical": { "score": 7, - "reason": "QA PASS at every LOD; 126k/52k/8.6k tris (hair ribbons are the bulk); 2 draws.", + "reason": "QA PASS at every LOD; 126.4k / 51.6k / 8.2k tris (hair ribbons are the bulk); 2 draws.", "fixes": [ "hair cards for the ribbons" ] } }, "renders": { - "hero.png": "8a9c16c6dd5f50c55309b65a9037b88c52d95d43a87a13120c44ec80f5edb1ad", - "sheet.png": "526f8cda813ceb9a4c7983cd0a1bddf1033cbafc56e05793d132839a49296da4", - "raking.png": "f7aa18e74ff205feedeabe21baf8d7ea952d7bbc7625999b6e2ecc9d6018d0be", - "raking_detail.png": "de183823c3228a52a9b526f9d7ba2ee914692ae3cab3cdc7801fd07f3f4411c4", - "lods.png": "6b7eb34321c70c297c602e2c99305d3a750a66ff7c04cd49534a54c6f35ad137", - "silhouette.png": "25e73964f5d2e2581628bdec5ddc6c8d1fde52551ceb7f12e96b1acfb2a3c7df", - "value.png": "0acb967cfeab57593e2bc433a959b0484be162f27306bcb7d64ec261b1cfeabe", - "detail.png": "83cd9a9d49cc65089508b7ffadee9b8a0dbbd7e18ec175f79fb4804996a43448" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.885 - }, - { - "view": "side", - "iou": 0.906 - }, - { - "view": "3/4", - "iou": 0.911 - } - ], - "focal": { - "peakPart": "hair", - "peak": [ - 379, - 101 - ], - "focalShare": 0.473, - "focalArea": 0.659, - "top": [ - { - "part": "skin", - "share": 0.473 - }, - { - "part": "hair", - "share": 0.334 - }, - { - "part": "plinth", - "share": 0.193 - } - ], - "declared": [ - "skin" - ] - }, - "rest": { - "share": 0.59, - "spread": 0.657, - "min": 0.25 - }, - "grounded": { - "rise": 0.252, - "bands": [ - 0.366, - 0.654, - 0.571, - 0.58, - 0.655 - ] - }, - "shape": { - "acuteShare": 0.111, - "straightShare": 0.578, - "taper": 0.213, - "shape": "stable" - } + "hero.png": "1b5068144ca238385628a8affed826cbf1e55c0db1446c2e7750da0789e77231", + "sheet.png": "f5e27519d87697763e7e8898864decbeb389b5cbdc3220a488409cd2195837a7" } } diff --git a/recipes/face_demo/out/face_goblin/critique.json b/recipes/face_demo/out/face_goblin/critique.json index c11de7b5..cf5e2260 100644 --- a/recipes/face_demo/out/face_goblin/critique.json +++ b/recipes/face_demo/out/face_goblin/critique.json @@ -1,130 +1,63 @@ { "asset": "face_goblin", "reviewer": "agent", - "overall": 6, + "overall": 5.6, "scores": { "silhouette": { "score": 7, - "reason": "Big vault, long nose, receding chin and 40°-swung pointed ears read as a goblin from front and side; the cleanest silhouette of the five.", + "reason": "Big vault, long nose, receding chin and swung pointed ears read as a goblin from front and side; the cleanest silhouette of the five.", "fixes": [ - "hook the nose tip more (hook 0.08 H)", - "notches on the ear rims" + "hook the nose tip more and notch the ear rims" ] }, "proportions": { "score": 7, - "reason": "Head ×0.84, vault ×1.16, nose ×2.6 projection hold; eye spacing 57.9 vs 59.0 mm. The eyes are human-sized; goblins usually read better with larger eyes.", + "reason": "Head, vault and nose projection hold; the eyes are human-sized, where goblins read better with larger eyes.", "fixes": [ "eyeball ×1.3 with a matching socket landmark scale" ] }, "materials": { "score": 4, - "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets, curvature cavity, mottle) separate by hue but not by roughness or SSS; hair is the same material as skin, so it reads as sculpted clay; eyes are the only glossy part (cornea clearcoat).", + "reason": "Vertex colour on the plain strip: green skin zones separate by hue but not by roughness or subsurface; it reads as matte painted clay. No hair.", "fixes": [ - "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen (material-artist)", - "darker hair roots and per-lock value variation along the lock (root → tip colour is set but too subtle)" + "a skin strip with subsurface-like roughness/sheen", + "glossy eyes and wet lips" ] }, "detail": { "score": 5, - "reason": "Primary and secondary forms come from a real anatomical base (eyelids, nasolabial area, lips, ear helix/antihelix/concha, clavicles, sternocleidomastoid) and read cleanly under the raking light, but there is no tertiary detail: no pores, wrinkles, lid creases or brow hairs (brows are painted bands), and the SDF hair locks are uniform grooves.", + "reason": "Primary and secondary forms come from a real anatomical base (lids, nasolabial area, lips, ear helix, clavicles, sternocleidomastoid), but there is no tertiary layer (pores, wrinkles) and the hair is evenly spaced grooves.", "fixes": [ - "LOD0 normal-scale wrinkle/pore displacement on the subdivided base driven by age", - "hair: varied lock widths and a parting instead of evenly spaced grooves; hair cards (planned)" + "LOD0 wrinkle/pore displacement on the subdivided base driven by age", + "varied lock widths and a parting instead of evenly spaced grooves" ] }, "wear": { "score": 3, "reason": "Pristine: seeded asymmetry and age/weight knobs are the only story; no scars, stubble, blemishes, sun-darkened skin or dirt.", "fixes": [ - "seeded scars/blemish decals in vertex colour (orc, dwarf)", - "stubble shadow on the human/dwarf jaw", - "age-driven darkening under the eyes" + "seeded scar/blemish marks in vertex colour", + "stubble shadow on the jaw and age-driven darkening under the eyes" ] }, "lod": { "score": 5, - "reason": "LOD2 loses most of each ear (thin sheets under the 1.4 cm cell) and the nose softens: the silhouette changes at distance.", + "reason": "LOD2 loses most of each ear (thin sheets under the 1.4 cm cell) and the face goes blotchy: the silhouette changes at distance.", "fixes": [ "authored LOD2 topology, or keep the ear shells from LOD1" ] }, "technical": { "score": 8, - "reason": "QA PASS at every LOD; 49k/13k/5.9k tris (no hair) — triangles go to the face; 2 draws.", + "reason": "QA PASS at every LOD; 48.9k / 13k / 5.7k tris (no hair); 2 draws.", "fixes": [ "drop the teeth at LOD1 (barely visible)" ] } }, "renders": { - "hero.png": "d18db0d865a84bd4ed01450e1059ad15001694192c422a1580c7fa0f7e9f155b", - "sheet.png": "cc270753ce8f7cb5548a5352c8e6e9e48852cc31ec9dbf783c0827449a0a9aa5", - "raking.png": "5e492fdc136dbcb49257f31f6cc85152dca6cabd49d4e1b9b36d538d83a13162", - "raking_detail.png": "d0b19e27d18f5d139101843fa8c672372fa3f32bfcef94c6958b1b2b1e622cf1", - "lods.png": "e3500fca249073ac09e71138da132a6ecd9264bae4d8bc39a13f3ffd46827842", - "silhouette.png": "b9ef177ea01c182583607d5765c0896af8a3c75b5bf44bc2e66cbe1ab58d4074", - "value.png": "5d1c31ff2a69dfaa418d5dcc0250ca174f6771d340b86c65be8221f5046fe7cb", - "detail.png": "7a3a6ce0965ca33d53c19bfbfb317e4cf043ad6c3b1a7a128bba953a9a81d382" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.972 - }, - { - "view": "side", - "iou": 0.969 - }, - { - "view": "3/4", - "iou": 0.962 - } - ], - "focal": { - "peakPart": "plinth", - "peak": [ - 396, - 360 - ], - "focalShare": 0.668, - "focalArea": 0.867, - "top": [ - { - "part": "skin", - "share": 0.668 - }, - { - "part": "plinth", - "share": 0.332 - } - ], - "declared": [ - "skin" - ] - }, - "rest": { - "share": 0.74, - "spread": 0.533, - "min": 0.25 - }, - "grounded": { - "rise": 0.328, - "bands": [ - 0.307, - 0.556, - 0.583, - 0.505, - 0.661 - ] - }, - "shape": { - "acuteShare": 0.217, - "straightShare": 0.456, - "taper": 0.138, - "shape": "friendly" - } + "hero.png": "c9f9319fedc7da46b3233cc80ad850f28f2eb00a5491a630021d5d4b54af84ad", + "sheet.png": "2fdad636f8fee476644842ea6fa7f185ded374f1ef9a0ea402af2a1facc44d18" } } diff --git a/recipes/face_demo/out/face_human/critique.json b/recipes/face_demo/out/face_human/critique.json index 20d77eda..7f852831 100644 --- a/recipes/face_demo/out/face_human/critique.json +++ b/recipes/face_demo/out/face_human/critique.json @@ -1,135 +1,64 @@ { "asset": "face_human", "reviewer": "agent", - "overall": 5, + "overall": 5.3, "scores": { "silhouette": { "score": 6, - "reason": "Reads as a human bust from front and side (IoU LOD2 ≥ 0.9); the outline is a calm classical bust. The helmet-like swept hair with a hard front edge is the weakest shape; nothing in the silhouette is characterful.", + "reason": "Reads as a human bust from front and side; the grooved helmet hair with a hard front edge is the weakest shape and nothing in the outline is characterful.", "fixes": [ - "give the hair volume over the crown and break the outline with a few lifted locks", - "a softer, lower front hairline" + "volume over the crown and a few lifted locks; a softer, lower front hairline" ] }, "proportions": { "score": 7, - "reason": "The base is a measured human (MakeHuman hm08, CC0): eye spacing 61.2 mm vs 61.6 spec (QA measures ±2 %), natural thirds; the face reads young and androgynous because the base is gender-neutral.", + "reason": "Measured human base (MakeHuman, CC0); eye spacing on spec, natural thirds; reads young and androgynous.", "fixes": [ - "add a masculine/feminine axis from MakeHuman's CC0 targets as a preset knob" + "a masculine/feminine axis from MakeHuman's CC0 targets as a preset knob" ] }, "materials": { "score": 4, - "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets, curvature cavity, mottle) separate by hue but not by roughness or SSS; hair is the same material as skin, so it reads as sculpted clay; eyes are the only glossy part (cornea clearcoat).", + "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets) separate by hue but not by roughness or subsurface; the hair is the same material as the skin, so it reads as sculpted clay; there is a pale band of bare skin under the hair helmet edge at the hairline in hero.png.", "fixes": [ - "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen (material-artist)", - "darker hair roots and per-lock value variation along the lock (root → tip colour is set but too subtle)" + "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen", + "darker hair roots and per-lock value variation along each lock", + "close the hairline: extend the hair mass to the skin (or tint the scalp under it to the root colour) so no pale band shows" ] }, "detail": { "score": 5, - "reason": "Primary and secondary forms come from a real anatomical base (eyelids, nasolabial area, lips, ear helix/antihelix/concha, clavicles, sternocleidomastoid) and read cleanly under the raking light, but there is no tertiary detail: no pores, wrinkles, lid creases or brow hairs (brows are painted bands), and the SDF hair locks are uniform grooves.", + "reason": "Primary and secondary forms come from a real anatomical base (lids, nasolabial area, lips, ear helix, clavicles, sternocleidomastoid), but there is no tertiary layer (pores, wrinkles) and the hair is evenly spaced grooves.", "fixes": [ - "LOD0 normal-scale wrinkle/pore displacement on the subdivided base driven by age", - "hair: varied lock widths and a parting instead of evenly spaced grooves; hair cards (planned)" + "LOD0 wrinkle/pore displacement on the subdivided base driven by age", + "varied lock widths and a parting instead of evenly spaced grooves" ] }, "wear": { "score": 3, "reason": "Pristine: seeded asymmetry and age/weight knobs are the only story; no scars, stubble, blemishes, sun-darkened skin or dirt.", "fixes": [ - "seeded scars/blemish decals in vertex colour (orc, dwarf)", - "stubble shadow on the human/dwarf jaw", - "age-driven darkening under the eyes" + "seeded scar/blemish marks in vertex colour", + "stubble shadow on the jaw and age-driven darkening under the eyes" ] }, "lod": { - "score": 6, - "reason": "LOD2 (1.4 cm re-mesh) keeps the outline and colour blocks; the face goes soft and the ears thin out.", + "score": 5, + "reason": "LOD2 keeps the outline and colour blocks, but the face breaks into blotchy patches and the bust edge is ragged in sheet.png.", "fixes": [ - "LOD2 from a coarser authored topology (MakeHuman proxy, CC0) instead of a distance-field re-mesh" + "LOD2 from a coarser authored topology (MakeHuman proxy) instead of a distance-field re-mesh" ] }, "technical": { "score": 7, - "reason": "QA PASS at every LOD, connections clean (anatomical folds declared from landmarks); 96k/37k/8.5k tris; 2 draws; but LOD0 spends ~45 % on SDF hair and each LOD0 build takes ~20 s.", + "reason": "QA PASS at every LOD; 96.2k / 36.7k / 8.2k tris; 2 draws; ~45% of LOD0 is SDF hair.", "fixes": [ - "cheaper hair (cards or a coarser sculpt cell)", - "cache the scalp field between LODs" + "cheaper hair (cards or a coarser sculpt cell)" ] } }, "renders": { - "hero.png": "b73b9db25abaa67210a84bc6837ad5d7dc6041aaba0f84fd7435ae11e98e8ab2", - "sheet.png": "0796fa4ac82bdbb9b79eb4ac81b9c8368334bd459036409514fb1e5a358312a3", - "raking.png": "d2ea31e6a065e897fa3802d69b3b6aa062a3c85874483b939203221272c1835b", - "raking_detail.png": "6803d11d1978c4d62ddad84f59c608487474ba00f2e9199e2277f9d21ceb0819", - "lods.png": "235780bd377dc7d3ec08ab5aba5aa260eb5b8431310e9f2fb5c92a2d7972be85", - "silhouette.png": "dfe19918b0659c9cca76cc2843d3dcde2a3c6323a2f9955bf465e498dde7e220", - "value.png": "e7254a1191708750e28608d9efe605e0cf3aa1392c2ddc5ae732cc6484dbfa00", - "detail.png": "55d1ca293f0200a8bbfedfee535d6a6bb8a373fdbf5dcb7f5bc771f5a30e7982" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.978 - }, - { - "view": "side", - "iou": 0.965 - }, - { - "view": "3/4", - "iou": 0.958 - } - ], - "focal": { - "peakPart": "skin", - "peak": [ - 391, - 138 - ], - "focalShare": 0.692, - "focalArea": 0.698, - "top": [ - { - "part": "skin", - "share": 0.692 - }, - { - "part": "plinth", - "share": 0.206 - }, - { - "part": "hair", - "share": 0.102 - } - ], - "declared": [ - "skin" - ] - }, - "rest": { - "share": 0.661, - "spread": 0.631, - "min": 0.25 - }, - "grounded": { - "rise": -0.003, - "bands": [ - 0.349, - 0.594, - 0.607, - 0.516, - 0.385 - ] - }, - "shape": { - "acuteShare": 0.042, - "straightShare": 0.502, - "taper": 0.239, - "shape": "stable" - } + "hero.png": "e55b0a4070fe1dabd881c3e6796923e7e37c887442fc6d715ba8b44e067ca991", + "sheet.png": "6e2d2c008d02a7543b82c73ebc7bb1e82d96430f28e1d106188248c45b420d77" } } diff --git a/recipes/face_demo/out/face_orc/critique.json b/recipes/face_demo/out/face_orc/critique.json index 2d71559a..c097ee07 100644 --- a/recipes/face_demo/out/face_orc/critique.json +++ b/recipes/face_demo/out/face_orc/critique.json @@ -1,135 +1,63 @@ { "asset": "face_orc", "reviewer": "agent", - "overall": 5, + "overall": 5.1, "scores": { "silhouette": { "score": 6, - "reason": "Heavy jaw (×1.4), brow shelf, thick neck and trapezius, crest and tusks read as an orc from the front and 3/4; from behind it is a bald green human with a crest.", + "reason": "Heavy jaw, brow shelf, thick neck, crest and tusks read as an orc from the front and 3/4; from behind it is a bald green human with a crest.", "fixes": [ - "push the trapezius higher and the head forward (hunched stance)", - "bigger, wider-set tusks and a heavier crest" + "push the trapezius higher and the head forward (hunched stance); bigger, wider-set tusks" ] }, "proportions": { "score": 6, - "reason": "Declared ratios hold (jaw ×1.4 lands on its landmark exactly, eye spacing 72.0 mm vs 73.1 spec); the face still reads too human between the cheekbones — nose ×1.8 wide but the mid-face is not flattened enough.", + "reason": "Declared ratios hold (jaw, eye spacing); the mid-face still reads too human between the cheekbones.", "fixes": [ - "flatten the nose bridge further (bridge −0.035 H) and widen the nostril flare", - "lower the eyes deeper under the brow shelf" + "flatten the nose bridge further and widen the nostril flare; lower the eyes under the brow shelf" ] }, "materials": { "score": 4, - "reason": "Vertex colour on the plain strip: skin zones (flush, lips, cool sockets, curvature cavity, mottle) separate by hue but not by roughness or SSS; hair is the same material as skin, so it reads as sculpted clay; eyes are the only glossy part (cornea clearcoat).", + "reason": "Vertex colour on the plain strip: green skin zones separate by hue but not by roughness or subsurface; the crest is the same material as the skin, so it reads as sculpted clay; tusks separate by hue only.", "fixes": [ - "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen (material-artist)", - "darker hair roots and per-lock value variation along the lock (root → tip colour is set but too subtle)" + "a skin strip with subsurface-like roughness/sheen and a hair strip with anisotropic sheen for the crest", + "glossier, slightly yellowed ivory on the tusks with a darker root" ] }, "detail": { "score": 5, - "reason": "Primary and secondary forms come from a real anatomical base (eyelids, nasolabial area, lips, ear helix/antihelix/concha, clavicles, sternocleidomastoid) and read cleanly under the raking light, but there is no tertiary detail: no pores, wrinkles, lid creases or brow hairs (brows are painted bands), and the SDF hair locks are uniform grooves.", + "reason": "Primary and secondary forms come from a real anatomical base (lids, nasolabial area, lips, ear helix, clavicles, sternocleidomastoid), but there is no tertiary layer (pores, wrinkles) and the hair is evenly spaced grooves.", "fixes": [ - "LOD0 normal-scale wrinkle/pore displacement on the subdivided base driven by age", - "hair: varied lock widths and a parting instead of evenly spaced grooves; hair cards (planned)" + "LOD0 wrinkle/pore displacement on the subdivided base driven by age", + "varied lock widths and a parting instead of evenly spaced grooves" ] }, "wear": { "score": 3, "reason": "Pristine: seeded asymmetry and age/weight knobs are the only story; no scars, stubble, blemishes, sun-darkened skin or dirt.", "fixes": [ - "seeded scars/blemish decals in vertex colour (orc, dwarf)", - "stubble shadow on the human/dwarf jaw", - "age-driven darkening under the eyes" + "seeded scar/blemish marks in vertex colour", + "stubble shadow on the jaw and age-driven darkening under the eyes" ] }, "lod": { - "score": 7, - "reason": "LOD2 keeps the crest, jaw and tusks (IoU front 0.98); face softens.", + "score": 5, + "reason": "LOD2 keeps the crest, jaw and outline, but the face in sheet.png breaks into dark blotches around the eyes and mouth and the tusks fragment.", "fixes": [ - "authored low-poly topology for LOD2" + "authored low-poly topology for LOD2; keep the tusks as separate cones at LOD2" ] }, "technical": { "score": 7, - "reason": "QA PASS at every LOD; 84k/30k/8.3k tris; teeth declared `rest` (gums not modelled), tusks rooted in the jaw.", + "reason": "QA PASS at every LOD; 83.6k / 29.5k / 8.1k tris; teeth declared rest (no gums), tusks rooted in the jaw.", "fixes": [ - "model a gum strip so the teeth can grow (`grown`) instead of rest" + "model a gum strip so the teeth can be grown instead of rest" ] } }, "renders": { - "hero.png": "3ce1bd706e71b126fc99563f2cf51b76066a9840a6291b2fc5e3193e66cc9ce0", - "sheet.png": "76d2fc4c9129e4a97147d046635ce1f346902560d9d6d76d377977e642f6045c", - "raking.png": "99fdae6e8243c93cf35f366088fe0c797e0138a157d964662a6acc565b8eb354", - "raking_detail.png": "04ca74d156da5bca589a1961c0f71c18776a87051d71acc38a22a7c573f952e6", - "lods.png": "ee3b2030d6e39d40ff1dadc0546066d7f63a52d5b33323cdf050ae1a2d12aba9", - "silhouette.png": "64335e56a1f32ad78742fa0f80ca13072ca0b5e2339fd74ee93d6a87bdfb25fe", - "value.png": "8b28bd10967ae988d082db8c689eb96ed882508fc2fff328b2605e753f9083f1", - "detail.png": "4a7063baf4fa2808cb428b876731fb6251cd81a879a5201af0a2fddd104ef8a8" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.978 - }, - { - "view": "side", - "iou": 0.945 - }, - { - "view": "3/4", - "iou": 0.955 - } - ], - "focal": { - "peakPart": "skin", - "peak": [ - 427, - 92 - ], - "focalShare": 0.688, - "focalArea": 0.822, - "top": [ - { - "part": "skin", - "share": 0.688 - }, - { - "part": "hair", - "share": 0.218 - }, - { - "part": "plinth", - "share": 0.053 - } - ], - "declared": [ - "skin" - ] - }, - "rest": { - "share": 0.719, - "spread": 0.606, - "min": 0.25 - }, - "grounded": { - "rise": 0.142, - "bands": [ - 0.354, - 0.494, - 0.546, - 0.471, - 0.507 - ] - }, - "shape": { - "acuteShare": 0.1, - "straightShare": 0.507, - "taper": 0.307, - "shape": "stable" - } + "hero.png": "98de887b11a02b38fb88ae0b7efac360a28182836b43a1eb958e89365c1b3140", + "sheet.png": "e12fd823b598a6be621839e1f9b65788eef59b2dc2a081831ab4198f9c353e8d" } } diff --git a/recipes/finish_demo/out/finish_demo/critique.json b/recipes/finish_demo/out/finish_demo/critique.json new file mode 100644 index 00000000..1edc10c8 --- /dev/null +++ b/recipes/finish_demo/out/finish_demo/critique.json @@ -0,0 +1,62 @@ +{ + "asset": "finish_demo", + "reviewer": "agent", + "overall": 6.1, + "scores": { + "silhouette": { + "score": 7, + "reason": "EUR pallet with its blocks, a framed crate with a diagonal brace and a bulging cask with hoops: each prop names itself. The cask bulge is visibly faceted (12–16 sides) in sheet.png.", + "fixes": [ + "raise the cask segment count at LOD0 (24+) so the bulge reads round" + ] + }, + "proportions": { + "score": 8, + "reason": "1.2 × 0.8 m EUR pallet at 144 mm high; crate and cask sizes plausible against the 0.1 m grid.", + "fixes": [ + "check the cask against a 225 l barrique (≈ 0.95 m tall, 0.7 m belly)" + ] + }, + "materials": { + "score": 5, + "reason": "hero.png shows wood and rusted hoops, but the crate and cask textures are low-resolution and blurry (blocky pixels on the crate panels); sheet.png renders the crate and cask as untextured white, so the sheet does not show what the hero shows.", + "fixes": [ + "give the crate and cask a wood colour in vertex paint (not white) so they read in every render path", + "higher-resolution finish albedo (or the trim sheet's wood strips) for the crate and cask" + ] + }, + "detail": { + "score": 5, + "reason": "3.2k LOD0 triangles: pallet boards and blocks, crate frame and brace, cask hoops. No nails, no stave seams as geometry, no bung hole, no crate lid boards.", + "fixes": [ + "stave seams and a bung on the cask", + "nail heads on the pallet deck boards and crate frame" + ] + }, + "wear": { + "score": 6, + "reason": "The finish layer adds rust on the hoops and stains; the pallet deck is clean and the code-built props carry no wear of their own.", + "fixes": [ + "grime and scuffs on the pallet deck and block faces" + ] + }, + "lod": { + "score": 5, + "reason": "LOD2 (284) turns the pallet into a solid slab with no gaps and the crate into a plain box: the outline changes at the switch.", + "fixes": [ + "keep the pallet's three bearer blocks and the gaps under the deck at LOD2" + ] + }, + "technical": { + "score": 7, + "reason": "QA PASS; 3 172 / 1 132 / 284, one draw. Tiny and cheap, but the sheet and hero disagree on the materials.", + "fixes": [ + "make the contact sheet render the finished textures (or the vertex colour carry them) so both views agree" + ] + } + }, + "renders": { + "hero.png": "ea71927d5d4f1cd40a8ee53ea1d41f56cfbd3c37a6bfa01119710dcd57490cb7", + "sheet.png": "480db2b0bf05ee55f850aadfc3c58d811527a9fd57e8006dba3fdd12a5e4e17c" + } +} diff --git a/recipes/fox/out/fox/critique.json b/recipes/fox/out/fox/critique.json new file mode 100644 index 00000000..fbebedee --- /dev/null +++ b/recipes/fox/out/fox/critique.json @@ -0,0 +1,64 @@ +{ + "asset": "fox", + "reviewer": "agent", + "overall": 6, + "scores": { + "silhouette": { + "score": 7, + "reason": "Pointed ears, long muzzle, black stockings and a white-tipped brush read as a red fox. In hero.png the body is deep and the legs long, so it reads closer to a coyote; the brush looks short from the 3/4.", + "fixes": [ + "make the brush fuller (larger radius) so it reads as a fox tail from every view", + "shorten the legs ~8% and slim the chest" + ] + }, + "proportions": { + "score": 6, + "reason": "1.1 × 0.2 × 0.5 m holds the sourced head-body, tail and shoulder height, but the trunk is deep, the ears smallish and the hind thighs pale like the belly.", + "fixes": [ + "enlarge the ears ~15%", + "colour the hind thighs red; the cream belly zone climbs too high" + ] + }, + "materials": { + "score": 6, + "reason": "Orange coat, cream belly and black stockings separate; the coat reads as painted vertex colour with sparse fur strands sticking out rather than a dense coat.", + "fixes": [ + "denser, shorter fur cards on the body so stray strands stop reading as hairs on a smooth skin", + "glossy nose leather and eyes" + ] + }, + "detail": { + "score": 6, + "reason": "Eyes, nose, toes, ruff and brush tip; face planes soft, no whiskers.", + "fixes": [ + "whiskers as thin tubes", + "sharper cheek and brow planes" + ] + }, + "wear": { + "score": 4, + "reason": "Clean: no mud, no wet fur, no snow on the coat.", + "fixes": [ + "mud on the paws and snow dusting on the back (grime with the ground mask)" + ] + }, + "lod": { + "score": 6, + "reason": "LOD2 (2.7k) keeps the outline but the legs become beige tubes with black only at the feet, so the stocking colour block shifts at distance.", + "fixes": [ + "paint the LOD2 legs black from the elbow down to match LOD0" + ] + }, + "technical": { + "score": 7, + "reason": "QA PASS; 152.1k / 27.3k / 2.7k triangles; 2 draws. LOD0 is heavy for a 6 kg animal.", + "fixes": [ + "cap LOD0 near 80k by coarsening the fur-card density" + ] + } + }, + "renders": { + "hero.png": "7ccd43bf8754fb8d7e8d04fffcd01b5aa03201fb45c9da72c4c46da28af724ca", + "sheet.png": "36df86d87dbf72e97537cc71506e97cfd3934904cb780b1a284d8e5badbcca9c" + } +} diff --git a/recipes/goblin_head/out/goblin_head/critique.json b/recipes/goblin_head/out/goblin_head/critique.json index 977807f8..7bd27ef2 100644 --- a/recipes/goblin_head/out/goblin_head/critique.json +++ b/recipes/goblin_head/out/goblin_head/critique.json @@ -1,133 +1,67 @@ { "asset": "goblin_head", "reviewer": "agent", - "overall": 6.0, + "overall": 5.6, "scores": { "silhouette": { "score": 7, - "reason": "Huge notched ears, hooked nose and dome read as a goblin; ear lobe balls removed; neck is a neck, not a pole.", + "reason": "Long pointed ears, hooked nose, bald dome with grey wisps and a thin neck read as a goblin in FRONT and SIDE; the shoulders are two round brown balloons with no garment shape, so the bust outline below the neck is generic.", "fixes": [ - "a stronger hunch in the side view" + "give the vest a cut shoulder seam and collar edge in the side profile instead of the inflated shoulder ellipsoids", + "a stronger forward hunch of the neck in the SIDE view" ] }, "proportions": { "score": 6, - "reason": "Neck 0.51 × head width (declared skinny range), SCM and windpipe read; mouth area still pursed and the grin reads weakly.", + "reason": "Head, ears and nose sit well; the neck is long and pole-straight in SIDE, and the shoulder masses are oversized against the narrow skull (wider than the ear span).", "fixes": [ - "widen the grin and show the teeth row", - "shorten the vault 5 %" + "narrow the shoulder masses ~15 % and drop them 2 cm so the neck shortens", + "widen the grin and show the teeth row (mouth reads pursed)" ] }, "materials": { "score": 5, - "reason": "Skin, shirt, vest, hood, scarf separate by hue; leather and linen share roughness.", + "reason": "Skin, vest, shirt and scarf separate by hue only; vest leather, linen shirt and wool scarf share one soft sheen; the scarf reads as red rubber tubing with a ball knot.", "fixes": [ - "leather strip on the vest", - "wool on the scarf" + "leather strip with grain and edge darkening on the vest", + "wool/knit strip and fraying on the scarf; flatten its round-tube section into a folded band" ] }, "detail": { "score": 6, - "reason": "Warts, brow creases, lids, wisps, hemmed vest, soft hood roll; ears alias along the rim.", + "reason": "Warts, brow creases, lids, ear notches and hair wisps hold at third person; vest and shirt are smooth masses with no seams, stitches or buttons.", "fixes": [ - "thicker ear rim at LOD0" + "seams, hem stitching and a toggle on the vest", + "thicker ear rim at LOD0 so the cartilage edge reads" ] }, "wear": { "score": 4, - "reason": "Some skin mottling; clothes clean.", + "reason": "Some skin mottling and red ear tips; clothes are clean and unpatched.", "fixes": [ - "patches and grime on the vest", - "torn scarf ends" + "patches, grime and frayed edges on the vest", + "torn, dirty scarf ends" ] }, "lod": { - "score": 7, - "reason": "LOD2 keeps ears and nose.", + "score": 4, + "reason": "FAR LOD2 in sheet.png loses both ears and the hair wisps: the head becomes a bald egg, so the goblin read pops away at distance.", "fixes": [ - "keep the wisps out at LOD2" + "keep the ears as a two-plane card (or a 6-segment cone) at LOD2", + "keep a single hair card at LOD2" ] }, "technical": { "score": 7, - "reason": "QA PASS honestly with 491 small crease regions (≤ 3 cm); socket joints pass; 80k / 22k / 4.2k tris.", + "reason": "QA clean on all LODs; 84k / 17k / 3.6k tris, heavy for a bust and spent largely on the smooth shoulder masses.", "fixes": [ + "cut the shoulder balloons to ~5k tris and spend on ear rims", "declare the warts from their seeded sizes instead of a 2× bound" ] } }, "renders": { - "hero.png": "65b94a562042ae0a46180599ca3e9e734922a1040c810a573185cc626eef9b76", - "sheet.png": "4b5b2dc86a4bc9f693c206ca9d56bbe012cd2a24d81877b26d5ec1a448dd4663", - "raking.png": "70303ba3f6f33ccc66ce125c457675a454fe26cfbd08d58631ab14286ebab412", - "raking_detail.png": "3093d840e5141d0c9aee4af2672d2ed1e294e3c1d43f19223d7159e051e3d641", - "lods.png": "9b0242b7d05c5126fe6fe4c109684a151563f193413e3e090b8217bdcaa72f06", - "silhouette.png": "26a7381d38df07a4210521264450557edcd154be8429bba2307a9a849d11048b", - "value.png": "afb393cd847cd1a6b61958f95eecc6f694a63d8d004d06bfb512c32a1f84a1f8", - "detail.png": "b449faf34245e2d4fe06e660ffdc0eaa73449330a395bd094d2ed9b919e9956f" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.949 - }, - { - "view": "side", - "iou": 0.936 - }, - { - "view": "3/4", - "iou": 0.937 - } - ], - "focal": { - "peakPart": "head", - "peak": [ - 473, - 160 - ], - "focalShare": 0.893, - "focalArea": 0.608, - "top": [ - { - "part": "head", - "share": 0.893 - }, - { - "part": "bust", - "share": 0.101 - }, - { - "part": "eye_r", - "share": 0.006 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.393, - "spread": 0.587, - "min": 0.25 - }, - "grounded": { - "rise": 0.524, - "bands": [ - 0.171, - 0.368, - 0.38, - 0.498, - 0.63 - ] - }, - "shape": { - "acuteShare": 0.307, - "straightShare": 0.428, - "taper": 0.329, - "shape": "aggressive", - "declared": "aggressive" - } + "hero.png": "09f1efa3b1f76db1583dc4f2f9a895031007d277e6ebae0ef80cf77348496b8b", + "sheet.png": "911334f7636d5e8234f385fb33ea78cd8b32c83f6c64f6f02eaf11699f5464b6" } } diff --git a/recipes/great_white/out/great_white/critique.json b/recipes/great_white/out/great_white/critique.json new file mode 100644 index 00000000..a6ba82d1 --- /dev/null +++ b/recipes/great_white/out/great_white/critique.json @@ -0,0 +1,64 @@ +{ + "asset": "great_white", + "reviewer": "agent", + "overall": 6.7, + "scores": { + "silhouette": { + "score": 8, + "reason": "Torpedo body, tall first dorsal, crescent caudal, pectorals and five gill slits read as a great white from every view.", + "fixes": [ + "a more lunate caudal with the lower lobe nearly as long as the upper" + ] + }, + "proportions": { + "score": 7, + "reason": "4.5 m TL with the sourced girth and dorsal position; the pectoral fins are paper-thin plates (a line in FRONT) and the snout is slightly pointed for a white shark.", + "fixes": [ + "give the pectorals a root thickness (~6 % of chord) tapering to the edge", + "blunt the snout cone ~10 %" + ] + }, + "materials": { + "score": 6, + "reason": "Countershading grey over white with a ragged boundary reads; skin is one uniform satin with fine speckle; the eye is a glossy black ball with no rim.", + "fixes": [ + "denticle micro-normal and slight sheen variation over the flank", + "an eyelid rim and darker pupil ring around the eye" + ] + }, + "detail": { + "score": 6, + "reason": "Gill slits, teeth row, pores on the snout, pelvic and second dorsal fins; no lateral line, claspers or fin-edge darkening.", + "fixes": [ + "dark pectoral tips and trailing edges", + "the lateral line and caudal keels" + ] + }, + "wear": { + "score": 6, + "reason": "Pink scratch scars on the back tell a fighting history.", + "fixes": [ + "a torn notch in the first dorsal", + "parasitic copepods on the fins" + ] + }, + "lod": { + "score": 7, + "reason": "FAR LOD2 keeps outline and countershading; the white boundary straightens to a hard edge.", + "fixes": [ + "keep the boundary wave at LOD2 by colouring along a jagged vertex row" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean; 61k / 7k / 680 tris across 7 parts; the LOD1 step is steep (8.7×).", + "fixes": [ + "raise LOD1 to ~15k so the fins and gills survive the first switch" + ] + } + }, + "renders": { + "hero.png": "34c7eca6469d52f85a074d99337971da7ab0c381d01320f98ffb12e85c835a06", + "sheet.png": "1dcbd70ce508d33a809d0016a27f61a976cad8ccbcbc754987fd6f924f53d959" + } +} diff --git a/recipes/human_head/out/human_head/critique.json b/recipes/human_head/out/human_head/critique.json index f676484a..7fce86d5 100644 --- a/recipes/human_head/out/human_head/critique.json +++ b/recipes/human_head/out/human_head/critique.json @@ -1,134 +1,66 @@ { "asset": "human_head", "reviewer": "agent", - "overall": 6.0, + "overall": 5.9, "scores": { "silhouette": { "score": 7, - "reason": "Reads as a bearded hero bust from front and side; swept hair and beard give a clear profile; LOD2 IoU ≥ 0.95 all views.", + "reason": "Swept hair, full beard and the profile nose read as a bearded man from FRONT and SIDE; the shoulders are two inflated red balloons that dominate the outline.", "fixes": [ - "a stronger hair silhouette at the crown (lift the front lock)" + "shape the mantle shoulders with a seam and a fold instead of round ellipsoids", + "lift the front lock at the crown for a stronger hair silhouette" ] }, "proportions": { "score": 6, - "reason": "Neck now 0.76 × head width with an SCM V, larynx and trapezius slope (checkNeck), mouth at the measured projection; but the brow/orbit area still reads heavy and the cheek-to-ear beard edge is a hard horizontal line.", + "reason": "Face, beard and ears sit well; the shoulder masses are very wide and bulbous for the head (deltoid width ≈ 3.3 head widths) and the beard top edge is a hard horizontal line.", "fixes": [ - "soften the beard top edge into the cheek (fade wider than 1.5 cm)", - "lift the brow ridge 2 mm and open the upper orbit" + "narrow and flatten the shoulder masses ~15 %", + "soften the beard top edge into the cheek over > 1.5 cm" ] }, "materials": { "score": 5, - "reason": "Skin, hair, tunic, mantle and brass separate by hue only; cloth and skin share one roughness.", + "reason": "Skin, hair, red mantle, purple tunic and brass studs separate by hue; mantle and tunic share skin-like roughness; brass studs read.", "fixes": [ - "leather/linen strips for collar and tunic", - "brass finish on the brooch" + "wool strip on the mantle and linen on the tunic with their own roughness", + "brass finish (metal, low roughness) on the brooch" ] }, "detail": { "score": 6, - "reason": "Lids, ears, beard and hair locks, tunic folds and a hemmed mantle hold up at third-person distance; face planes are smoother than a pro bust.", + "reason": "Lids, beard and hair locks, studs and a brooch hold up at third person; the mantle is a smooth balloon without folds or hem.", "fixes": [ - "nasolabial and lid folds at LOD0", - "stitching on the mantle hem" + "fold lines and a hemmed edge on the mantle", + "nasolabial and lid folds at LOD0" ] }, "wear": { "score": 4, - "reason": "Clean garments, no story beyond the painted tone.", + "reason": "Clean garments; nothing says where he has been.", "fixes": [ "grime at the mantle hem and collar", - "a scar or weathering on the face" + "a scar or sun weathering on the face" ] }, "lod": { "score": 7, - "reason": "LOD2 keeps outline and colour blocks; the face flattens to a mask at LOD2.", + "reason": "FAR LOD2 keeps outline, beard and the colour blocks; the face flattens and the tunic colours break into blocks.", "fixes": [ "keep the brow/nose mass at LOD2" ] }, "technical": { - "score": 7, - "reason": "QA PASS honestly: 578 crease regions, each ≤ 2.5 cm on a declared fold (no area regions); eye socket joints pass at LOD0; 71k / 25k / 3.8k tris.", + "score": 6, + "reason": "QA clean; 101k / 28k / 3.6k tris for a bust, a third of it on featureless mantle shoulders.", "fixes": [ + "cap LOD0 near 60k: decimate the mantle shoulders at source (fewer rings)", "eye socket joints at LOD1 need a coarser bore" ] } }, "renders": { - "hero.png": "3b69185c16f915ec9b422c2e9dc5e6c80c53a2ab0f3ef36f163fbb979dde12be", - "sheet.png": "b4666ad6acd3ea08f4081c7b032cd1fb3e09d3a9d00693433e2fa62c6339d5ec", - "raking.png": "8d9cabe2bc90bcb2b7da55304d757df6dd4007ced89bc6f1a48a9334807fb672", - "raking_detail.png": "b66a598d622fbd47513cf1af245a71840d46420cb1161ceb94a1912cdbb9e000", - "lods.png": "63848de129cd8b92542671f92041d16cc5736403b6eb46a4cc8c2e6017747ff1", - "silhouette.png": "d408b0f14866b336910a6ae4dc80780a1820c8650699cef1daa11c6b5d7eb932", - "value.png": "ef9241e41ee3980c0a639fa82ff295820507e77b1598720beb1d2200bfac12d8", - "detail.png": "63b936add0be0661f455a63927906561df36f1f93857e82c5e579a3b177f78cf" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.971 - }, - { - "view": "side", - "iou": 0.952 - }, - { - "view": "3/4", - "iou": 0.965 - } - ], - "focal": { - "peakPart": "head", - "peak": [ - 435, - 137 - ], - "focalShare": 0.967, - "focalArea": 0.499, - "top": [ - { - "part": "head", - "share": 0.967 - }, - { - "part": "brooch", - "share": 0.019 - }, - { - "part": "eye_r", - "share": 0.011 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.476, - "spread": 0.609, - "min": 0.25 - }, - "grounded": { - "rise": 0.282, - "bands": [ - 0.153, - 0.244, - 0.34, - 0.367, - 0.373 - ] - }, - "shape": { - "acuteShare": 0.03, - "straightShare": 0.535, - "taper": 0.237, - "shape": "stable", - "declared": "stable" - } + "hero.png": "65b6c79259fa32f3b0de585d01884ddf4ed5e294f96436498552caebe23cc031", + "sheet.png": "6f66b3808f49eb834e1d1f495722a9a7229bb30b6299c0fe9a00bf672ef7b5bc" } } diff --git a/recipes/hydra/out/hydra/critique.json b/recipes/hydra/out/hydra/critique.json new file mode 100644 index 00000000..81724ab5 --- /dev/null +++ b/recipes/hydra/out/hydra/critique.json @@ -0,0 +1,66 @@ +{ + "asset": "hydra", + "reviewer": "agent", + "overall": 5.9, + "scores": { + "silhouette": { + "score": 7, + "reason": "Five necks with spiked crests and horned heads on a sprawling four-legged body read as a hydra; one head hangs low in front of the chest in hero.png and muddles the leg silhouette.", + "fixes": [ + "raise the lowest neck so all five heads separate above the shoulders in the 3/4 view", + "larger dorsal spines along the back" + ] + }, + "proportions": { + "score": 6, + "reason": "Legs are thin and splayed for a 7.7 m animal; the trunk is small against five long necks and a long tail, and the chest has a pendulous blob between the forelegs in FRONT.", + "fixes": [ + "thicken the upper legs ~25 % and pull the feet under the body", + "flatten the chest bulge under the neck roots" + ] + }, + "materials": { + "score": 5, + "reason": "One olive skin with a lighter belly and paler belly scutes; horns, claws and teeth separate only by hue.", + "fixes": [ + "scale strip (or normal) on the back and a separate keratin roughness for horns and claws", + "darker dorsal saddle pattern to break the uniform olive" + ] + }, + "detail": { + "score": 6, + "reason": "Crests, horns, claws, teeth and belly scutes read; the skin between is smooth and uniform.", + "fixes": [ + "scale rows along the necks and plates on the back", + "gums and tongue in the open mouths" + ] + }, + "wear": { + "score": 4, + "reason": "No scars, broken horns or grime.", + "fixes": [ + "a stump (cauterised) or scarred neck, chipped horns", + "mud on the feet and belly" + ] + }, + "lod": { + "score": 7, + "reason": "FAR LOD2 keeps all five necks and the outline; heads simplify to wedges.", + "fixes": [ + "keep the horn pairs as cones at LOD2" + ] + }, + "technical": { + "score": 6, + "reason": "QA clean; 348k / 39k / 7.2k tris: LOD0 is very heavy for an in-game creature and drops 9× to LOD1.", + "fixes": [ + "lower neck and tail ring counts at LOD0 to ~120k total", + "spend saved triangles on the heads" + ] + } + }, + "renders": { + "hero.png": "e8df0660073e47fb822658ca2e7772759a08d7fb52438d0fc9e6434b93d71241", + "sheet.png": "61c1cd270f117eef07b30aab78a173f86412bcaf9d633e0391d743030e1f5e91" + } +} diff --git a/recipes/import_demo/out/import_demo/critique.json b/recipes/import_demo/out/import_demo/critique.json new file mode 100644 index 00000000..2e898b02 --- /dev/null +++ b/recipes/import_demo/out/import_demo/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "import_demo", + "reviewer": "agent", + "overall": 5.9, + "scores": { + "silhouette": { + "score": 6, + "reason": "A cropped-hair head on a bust reads; the shoulders are lumpy balloons and the neck ring stands off the neck in SIDE.", + "fixes": [ + "seat the neck ring on the neck surface (it overhangs at the back in SIDE)", + "shape the bust shoulders with a garment edge instead of round lumps" + ] + }, + "proportions": { + "score": 6, + "reason": "Head from the base mesh is human; the bust shoulders are very wide and the arms stumps bulge.", + "fixes": [ + "narrow the code-built bust shoulders ~15 %" + ] + }, + "materials": { + "score": 5, + "reason": "Skin, hair, blue tunic and brown shoulder cloth separate by hue; tunic and skin share one roughness; brass studs read.", + "fixes": [ + "give the tunic and the shoulder cloth their own cloth strips", + "roughness variation on the skin" + ] + }, + "detail": { + "score": 5, + "reason": "Scan face and brass studs; the bust has stains but no seams or folds.", + "fixes": [ + "seams and folds on the bust", + "a pore normal visible at hero distance" + ] + }, + "wear": { + "score": 5, + "reason": "Stains and darker patches on the tunic.", + "fixes": [ + "fray at the cloth edges and dirt in the folds" + ] + }, + "lod": { + "score": 7, + "reason": "FAR LOD2 keeps the head and colour blocks; the face becomes faceted.", + "fixes": [ + "raise the LOD2 face budget a little and keep eye sockets" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean; 46k / 18k / 3.3k tris; imported head simplified per LOD.", + "fixes": [ + "bake the scan pores to a normal map instead of vertex density" + ] + } + }, + "renders": { + "hero.png": "50210e05f6ab300e08b776db69763159712d8025aebc54bffb4a47b38b710a60", + "sheet.png": "1ba2aaab62635b73aae582f32129e3c93c48a860b593348cab4d61ce4a417611" + } +} diff --git a/recipes/joint_demo/out/joint_demo/critique.json b/recipes/joint_demo/out/joint_demo/critique.json new file mode 100644 index 00000000..3611cc63 --- /dev/null +++ b/recipes/joint_demo/out/joint_demo/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "joint_demo", + "reviewer": "agent", + "overall": 5.3, + "scores": { + "silhouette": { + "score": 6, + "reason": "An arm on a stand holding a sword reads as a joint demo; the blade is a needle-thin spike that vanishes edge-on in SIDE.", + "fixes": [ + "give the blade a real width (~4 cm) and fuller so it reads as a sword from the side" + ] + }, + "proportions": { + "score": 5, + "reason": "Arm length plausible; the hand is a featureless block fist and the blade is far too thin for its length.", + "fixes": [ + "model a hand with four finger rolls wrapped round the grip", + "blade width and taper from a real arming sword" + ] + }, + "materials": { + "score": 5, + "reason": "Skin-tone arm, grey steel blade, brass guard and pommel, black stand separate by hue; the blade does not read as polished metal.", + "fixes": [ + "metal finish with low roughness on the blade", + "leather wrap on the grip" + ] + }, + "detail": { + "score": 3, + "reason": "No knuckles, nails, wrist or elbow anatomy; plain guard and pommel.", + "fixes": [ + "elbow and wrist landmarks on the arm", + "a wrapped grip and shaped guard" + ] + }, + "wear": { + "score": 2, + "reason": "Spotless.", + "fixes": [ + "nicks along the blade edge and a worn grip" + ] + }, + "lod": { + "score": 8, + "reason": "FAR LOD2 is indistinguishable from LOD0 at that distance.", + "fixes": [ + "keep the guard bar at LOD2" + ] + }, + "technical": { + "score": 8, + "reason": "QA clean; 1.7k / 816 / 300 tris over 13 parts; joints check.", + "fixes": [ + "merge the static stand parts into one draw" + ] + } + }, + "renders": { + "hero.png": "f01b8fb9a57a33826057cf8cb62dc6f0b7ded30794923a0f27355017c14add47", + "sheet.png": "97e73b6c621b6ad0abb2c47e81737ffc9ce738acbdad3bfd3a2c63c1b42ca733" + } +} diff --git a/recipes/king_cobra/out/king_cobra/critique.json b/recipes/king_cobra/out/king_cobra/critique.json new file mode 100644 index 00000000..bcc2258d --- /dev/null +++ b/recipes/king_cobra/out/king_cobra/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "king_cobra", + "reviewer": "agent", + "overall": 5.9, + "scores": { + "silhouette": { + "score": 7, + "reason": "A raised forebody with a hood over a ground coil reads as a cobra; the hood is narrow and the coil loose.", + "fixes": [ + "wider, flatter hood ribs at the raised neck", + "tighter coil stack under the raised forebody" + ] + }, + "proportions": { + "score": 6, + "reason": "Raised height and coil size plausible for a 3–4 m snake; the body is near-constant girth with a thin tail and an abrupt neck.", + "fixes": [ + "taper girth from mid-body to the tail and swell the mid-body" + ] + }, + "materials": { + "score": 6, + "reason": "Dark scales with pale chevrons and cream ventral scutes read as scales; the chevron zigzag is closer to a viper pattern than the king cobra crossbands.", + "fixes": [ + "pale crossbands (chevrons pointing to the head) instead of zigzags", + "glossy scale roughness (low) on the dorsum" + ] + }, + "detail": { + "score": 6, + "reason": "Ventral scutes, head shields and tongue read; no eyes catchlight, no keeled scale normal.", + "fixes": [ + "visible eyes with a brille", + "scale normal map on the flanks" + ] + }, + "wear": { + "score": 4, + "reason": "Clean; no shed remnants or scars.", + "fixes": [ + "a dull pre-shed patch or a healed scar" + ] + }, + "lod": { + "score": 5, + "reason": "FAR LOD2 replaces the pattern with big black/cream blocks: the colour shifts at distance.", + "fixes": [ + "average the pattern to the dorsal grey-brown at LOD2 instead of sampling per band" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean; 62k / 3.8k / 584 tris in one part; the LOD0→LOD1 drop is 16× (a visible pop risk).", + "fixes": [ + "raise LOD1 to ~12k or lower LOD0 ring density along the straight neck" + ] + } + }, + "renders": { + "hero.png": "60a9d78fe54a8af71c95a41b46d384fd62051e2d27bcff3909477abf117ed1e2", + "sheet.png": "7d67e9a54340e61f366c7dab159d3d375ce31aecddd337f21238c0c7f550c6fb" + } +} diff --git a/recipes/loco_440/out/loco_440/critique.json b/recipes/loco_440/out/loco_440/critique.json new file mode 100644 index 00000000..b7266ad9 --- /dev/null +++ b/recipes/loco_440/out/loco_440/critique.json @@ -0,0 +1,65 @@ +{ + "asset": "loco_440", + "reviewer": "agent", + "overall": 6.3, + "scores": { + "silhouette": { + "score": 8, + "reason": "Diamond-less tall stack, domes, bell, headlamp, cowcatcher, cab and tender read as an 1880s 4-4-0 in every view.", + "fixes": [ + "flared balloon stack top a touch wider" + ] + }, + "proportions": { + "score": 7, + "reason": "4-4-0 wheel arrangement, drivers and boiler length read right; the tender is a plain box and the cab roof overhang is generous.", + "fixes": [ + "tender sides with flare and coping", + "check the cab roof overhang against the Jupiter replica photos" + ] + }, + "materials": { + "score": 6, + "reason": "Paint, brass, black iron and wood separate; the blue is a flat saturated plastic and the brass and black boiler share one sheen.", + "fixes": [ + "clearcoat paint with orange-peel roughness on the blue", + "polished brass (metal, low roughness) on the domes and bands" + ] + }, + "detail": { + "score": 7, + "reason": "Spoked drivers, rods, crossheads, bell, sand and steam domes, piping, lettered panels and sleepers; no rivets on the smokebox, no lining on the cab.", + "fixes": [ + "rivet rows on smokebox and firebox", + "pin-striping on the cab and tender panels" + ] + }, + "wear": { + "score": 3, + "reason": "Showroom-clean: no soot on the stack, no oil on the motion, no rust.", + "fixes": [ + "soot down the stack and smokebox front", + "oil and grime on the motion and wheels" + ] + }, + "lod": { + "score": 6, + "reason": "FAR LOD2 turns the spoked drivers into solid red blocks and the track into a white slab: colour and shape pop.", + "fixes": [ + "keep wheels as 8-sided discs with a dark hub at LOD2", + "keep the track colour (wood/steel) at LOD2" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean; 46k / 19k / 1.4k tris over 84 parts.", + "fixes": [ + "merge static body parts to cut draws; keep pivots only on wheels and rods" + ] + } + }, + "renders": { + "hero.png": "6eccdaedb5d39535800e6426da4d601325a86f766c8ada9f8c42b688e67ba1d7", + "sheet.png": "9742c51f21ec65c1d2237145fcc51e9f1705372c27760e134c785a2aa8261946" + } +} diff --git a/recipes/material_swatches/out/material_swatches/critique.json b/recipes/material_swatches/out/material_swatches/critique.json index bc7de42a..3874bc25 100644 --- a/recipes/material_swatches/out/material_swatches/critique.json +++ b/recipes/material_swatches/out/material_swatches/critique.json @@ -5,105 +5,56 @@ "scores": { "silhouette": { "score": 8, - "reason": "A stepped test-card stand: twelve rows of strip panels rising toward the back, the decal and foliage cells on the top row. Every panel is visible from the 3/4 view.", - "fixes": [] + "reason": "A stepped test-card stand: rows of strip panels rising to the back, decal and foliage cells on top; every panel visible from the 3/4 view.", + "fixes": [ + "separate the rows by a thin gap so the steps read in the SIDE view" + ] }, "proportions": { "score": 9, - "reason": "Each of the 42 strips is one true-scale repeat wide (stripRect(id).metres: 4 / 16 / 32 m) and its design height tall, so texel density can be judged directly against the 1.8 m figure.", - "fixes": [] + "reason": "Each strip one true-scale repeat wide, so texel density is judged directly.", + "fixes": [ + "add a 1.8 m figure at the front row as a scale reference" + ] }, "materials": { "score": 8, - "reason": "All 42 strips (classic, library, nature) render with their own albedo/normal/ORM: metals reflect, wood shows grain, cobbles and cliff show relief, glass is transparent, foliage cells cut out.", + "reason": "In hero.png the strips render with their own albedo/normal/ORM: wood grain, cobbles, cliff, grass, glass; in sheet.png most strips render flat grey.", "fixes": [ - "add a sphere per strip at LOD0 for curvature/roughness reading (docs/media/materials.png does this)" + "add a sphere per strip at LOD0 for curvature/roughness reading" ] }, "detail": { "score": 7, - "reason": "Labels in block letters on every plinth face; at hero distance the labels are too small to read for the long nature rows.", + "reason": "Labels on every plinth face; at hero distance the labels are unreadable.", "fixes": [ - "scale label pixel size with the row's panel width" + "scale label pixel size with the row panel width" ] }, "wear": { "score": 8, - "reason": "A test card is deliberately clean: weathering would contaminate the strip read. Weathering variants are shown by weather_demo.", - "fixes": [] + "reason": "A test card is clean by design; weathering variants live in weather_demo.", + "fixes": [ + "a worn variant row showing each strip with its grime pass" + ] }, "lod": { "score": 7, - "reason": "LOD1/LOD2 keep every panel (778 triangles) but drop labels.", - "fixes": [] + "reason": "LOD2 keeps every panel but drops labels.", + "fixes": [ + "keep row labels as one quad per label at LOD1" + ] }, "technical": { "score": 8, - "reason": "QA PASS, 27 288 / 778 / 778 (labels are the LOD0 cost), three draws (opaque, glass, cutout).", + "reason": "QA clean; 27k / 778 / 778 tris (labels are the LOD0 cost), 132 parts.", "fixes": [ "merge label pixel runs per word to cut LOD0 triangles" ] } }, "renders": { - "hero.png": "992f29b4b5983e14ab6f342d42138c17feff159a4146b5d09dc34a633a0bea98", - "sheet.png": "4cef5f87119c9a8e366340b8906db1dd9d5582ad9ee321e01fc6174f303fba49", - "raking.png": "b0ad06407e6698dd7d8e2b2bad8d90f6ee218d684514a74ca7f5c58454505def", - "raking_detail.png": "02eed6bf66e3e566e2b3cf1148484f730a1718b6864c2c19b1bf8ac568c53ee2", - "lods.png": "da7c27f1caf6c404ac3d3ba30cc2a27a1bd8881eb51d8a04d27c7d6222e772c3", - "silhouette.png": "e9ab1738e712b10663a6fb4c96dbcd0bcbee2666bd04f5fb0bfaa2a24a257263", - "value.png": "3d0f63c6d81fa00ff0039f701844643741ae8a1dcc17f2c45a6aebc236155fe0", - "detail.png": "23c4b3c851ff8ab5d028b31e1d94babc174ab6e3261eb01550aa1e6951d3fee6" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 1 - }, - { - "view": "side", - "iou": 0.999 - }, - { - "view": "3/4", - "iou": 1 - } - ], - "focal": { - "peakPart": "plinth_cells", - "peak": [ - 282, - 67 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "plinth_0", - "share": 0.24 - }, - { - "part": "plinth_cells", - "share": 0.155 - }, - { - "part": "strip_soil", - "share": 0.091 - } - ], - "declared": [] - }, - "rest": { - "share": 0.624, - "spread": 0.607, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.051, - "straightShare": 0.805, - "taper": 0.372, - "shape": "stable" - } + "hero.png": "578d0df2543cd61f6c51535190715a4bede445429bcee70a310e9c5d00cc6824", + "sheet.png": "c7f2c4ed35885c119fb73372ff5a95ac8a63939b99361522193431cb5e766074" } } diff --git a/recipes/mountain_range/out/mountain_range/critique.json b/recipes/mountain_range/out/mountain_range/critique.json index 02976393..1e1912c3 100644 --- a/recipes/mountain_range/out/mountain_range/critique.json +++ b/recipes/mountain_range/out/mountain_range/critique.json @@ -1,102 +1,62 @@ { "asset": "mountain_range", "reviewer": "agent", - "overall": 7.6, + "overall": 7.4, "scores": { "silhouette": { "score": 8, - "reason": "2 \u00d7 2 km alpine tile with eroded ridges, a glacial U-valley, peaks above the snowline; reads as mountains from any angle.", - "fixes": [] + "reason": "Snow-capped eroded peaks over a U-valley read as alpine mountains from every view.", + "fixes": [ + "sharper arête crests on the right ridge" + ] }, "proportions": { "score": 8, - "reason": "Relief 950 m over a 1 720 m floor, tree line 2 000 m, snow line 2 480 m, plausible for the Alps.", - "fixes": [] + "reason": "About 1 050 m relief on a 2 × 2 km tile, plausible alpine slopes.", + "fixes": [ + "a steeper headwall at the valley end (cirque)" + ] }, "materials": { - "score": 7, - "reason": "Ground strips chosen per triangle by slope and altitude: snow on gentle high ground, granite cliff on steep faces, gravel scree, grass meadows, moss on the forest floor; the cut sides show rock_strata beds. On 7.8 m cells the strips' V is squeezed per triangle, which draws fine banding on long slopes.", + "score": 6, + "reason": "Snow, rock, scree and meadow strips by slope/altitude read, but fine diagonal moiré banding covers every slope in hero.png and the snow edge is hard; the cut sides are plain brown with no strata.", "fixes": [ - "blend strips across triangle borders (vertex weights) instead of hard per-triangle switches", - "finer UV scale on the skirt sides" + "blend strips across triangle borders (vertex weights) instead of per-triangle switches", + "rock_strata beds visible on the skirt sides" ] }, "detail": { "score": 7, - "reason": "Hydraulic and thermal erosion channels, talus, ridges at 256\u00b2 cells; no trees or rocks scattered on the surface.", + "reason": "Erosion gullies and talus on the left peaks; the valley floor and the right slope are smooth.", "fixes": [ - "scatter tree impostors below the tree line and boulders on the scree" + "scatter tree impostors below the tree line and boulders on the scree", + "more erosion on the long smooth slopes" ] }, "wear": { "score": 7, - "reason": "Erosion gullies, scree fans and snow only on gentle slopes tell the weather; no avalanche paths or streams.", + "reason": "Erosion and snow on gentle slopes tell the weather.", "fixes": [ "a stream along the valley floor (water strip)" ] }, "lod": { "score": 8, - "reason": "Same grid at 256/128/64 cells; strips classified the same way at each LOD, so colour bands hold.", - "fixes": [] + "reason": "Same grid at 256/128/64 cells; FAR LOD2 holds outline and colour.", + "fixes": [ + "keep the snow line edge at LOD2 (it softens)" + ] }, "technical": { "score": 8, - "reason": "QA PASS all three seeds: 133 120 / 33 792 / 8 704, one draw.", - "fixes": [] + "reason": "QA clean; 133k / 34k / 8.7k tris, one draw.", + "fixes": [ + "drop skirt triangles that are never seen from above" + ] } }, "renders": { - "hero.png": "3cf0c5d356f03cd0f572ac7783da6ffd5971c8a36637508ea36e7aa27e970708", - "sheet.png": "07460bf69943e60f4229dbff902bd38b229a7d046e1de58675f0a321c85d39d8", - "raking.png": "5bea3e9737767048634058e2f9939f0c57540882c10285d5e6e77a5522dafd90", - "raking_detail.png": "7de5ecf321ca67064a23be6687096178be973266e85c5dde09de73cef932f85f", - "lods.png": "8ab7d9db2e30f848ba559b7190bdff9cb1105d12023757319a433ac9e204bf71", - "silhouette.png": "97f8ec9a7ff94f15d1df3a3a96aee3e083ad4e1126f08dfefef96a677a9af256", - "value.png": "1c27582cda5533cd74c8c9318015fbacf00538b7039731bd718e034a1b28e0af", - "detail.png": "77aaa70176e4c07c99da6963f3a57f51f79d4b6339fab7dfae1940249712c626" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.996 - }, - { - "view": "side", - "iou": 0.995 - }, - { - "view": "3/4", - "iou": 0.999 - } - ], - "focal": { - "peakPart": "terrain", - "peak": [ - 375, - 119 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "terrain", - "share": 1 - } - ], - "declared": [] - }, - "rest": { - "share": 0.732, - "spread": 0.669, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.12, - "straightShare": 0.726, - "taper": 0.451, - "shape": "stable" - } + "hero.png": "d6e3bbc9b9a66fcc3474ba6c9b94ac340f11b81da42932766ab440c85d58ae7c", + "sheet.png": "fcc2fa20169151863adcecceece3488b9cdd1a7b07bdfc69136f410639b69252" } } diff --git a/recipes/mountain_range/out/mountain_range_b/critique.json b/recipes/mountain_range/out/mountain_range_b/critique.json index a27bd8a2..bddfbeb3 100644 --- a/recipes/mountain_range/out/mountain_range_b/critique.json +++ b/recipes/mountain_range/out/mountain_range_b/critique.json @@ -1,102 +1,63 @@ { "asset": "mountain_range_b", "reviewer": "agent", - "overall": 7.6, + "overall": 6.9, "scores": { "silhouette": { - "score": 8, - "reason": "2 \u00d7 2 km alpine tile with eroded ridges, a glacial U-valley, peaks above the snowline; reads as mountains from any angle.", - "fixes": [] + "score": 7, + "reason": "A deep valley between peaks reads; the right summit is an isolated needle spike and the long slopes are smooth, so it reads less alpine than seed A.", + "fixes": [ + "clamp single-cell spikes (needle peak at the right edge)", + "more ridge branching on the long slopes" + ] }, "proportions": { - "score": 8, - "reason": "Relief 950 m over a 1 720 m floor, tree line 2 000 m, snow line 2 480 m, plausible for the Alps.", - "fixes": [] + "score": 7, + "reason": "About 1 060 m relief on 2 × 2 km; the needle peak is steeper than real rock holds.", + "fixes": [ + "limit max slope to ~60° outside cliff cells" + ] }, "materials": { - "score": 7, - "reason": "Ground strips chosen per triangle by slope and altitude: snow on gentle high ground, granite cliff on steep faces, gravel scree, grass meadows, moss on the forest floor; the cut sides show rock_strata beds. On 7.8 m cells the strips' V is squeezed per triangle, which draws fine banding on long slopes.", + "score": 6, + "reason": "Hero shows strong moiré banding on the long slopes and almost no snow; meadow green only in the valley.", "fixes": [ - "blend strips across triangle borders (vertex weights) instead of hard per-triangle switches", - "finer UV scale on the skirt sides" + "blend strips across triangle borders", + "lower the snow line for this seed or add snow on north faces" ] }, "detail": { - "score": 7, - "reason": "Hydraulic and thermal erosion channels, talus, ridges at 256\u00b2 cells; no trees or rocks scattered on the surface.", + "score": 6, + "reason": "Gullies on the left ridge; the large right slope is smooth.", "fixes": [ - "scatter tree impostors below the tree line and boulders on the scree" + "more erosion iterations on the smooth slopes", + "scree fans at the slope feet" ] }, "wear": { - "score": 7, - "reason": "Erosion gullies, scree fans and snow only on gentle slopes tell the weather; no avalanche paths or streams.", + "score": 6, + "reason": "Erosion only; no streams or avalanche tracks.", "fixes": [ "a stream along the valley floor (water strip)" ] }, "lod": { "score": 8, - "reason": "Same grid at 256/128/64 cells; strips classified the same way at each LOD, so colour bands hold.", - "fixes": [] + "reason": "FAR LOD2 holds outline and colour.", + "fixes": [ + "keep the needle peak from popping lower at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS all three seeds: 133 120 / 33 792 / 8 704, one draw.", - "fixes": [] + "reason": "QA clean; 133k / 34k / 8.7k tris, one draw.", + "fixes": [ + "drop skirt triangles that are never seen from above" + ] } }, "renders": { - "hero.png": "60a3b5bc3f463799bc2575bda8214143338f4b3446e861066c2bb787be19edd7", - "sheet.png": "f2605374df74792a10ff215979af1124afba5f46e8b2122b9ade2ffa1a8e5bc2", - "raking.png": "ce53a366a023db9a3b3b922908229d6c48324bcf06d6072057ba950b1fb9cf22", - "raking_detail.png": "d3b167f26baee28345d40c1db3bf5db0e41e22a83cd6927dff5a886c8219879c", - "lods.png": "41e92527e58eb61a0119c50507e52d92832e812b7e69dd9f2256a94e3543919c", - "silhouette.png": "6df1abc2e4b2f6905fcc02f2bf815e97678097d00b62a2f8c380732ac2667616", - "value.png": "bb6c067215c2385a787a135909ccb6e02c3db65938df3c855829893b15dc52f1", - "detail.png": "9a4bfd4d5f330760ef80ee5abff609a966ec333b6d4641725f6c1beaafeca3f5" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.994 - }, - { - "view": "side", - "iou": 0.996 - }, - { - "view": "3/4", - "iou": 0.997 - } - ], - "focal": { - "peakPart": "terrain", - "peak": [ - 280, - 153 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "terrain", - "share": 1 - } - ], - "declared": [] - }, - "rest": { - "share": 0.8, - "spread": 0.553, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.29, - "straightShare": 0.643, - "taper": 0.616, - "shape": "stable" - } + "hero.png": "e64f065f0c31942ae161cf21ab7a0a93c0101e8e72ee0a5c4e2ceedd071cc7a5", + "sheet.png": "2f2a9210bcad1bd0985ba8a69bfcf87734452d75f8090153bd9fbf12a518b4a3" } } diff --git a/recipes/mountain_range/out/mountain_range_c/critique.json b/recipes/mountain_range/out/mountain_range_c/critique.json index 9b3ddc1a..4a9f6d46 100644 --- a/recipes/mountain_range/out/mountain_range_c/critique.json +++ b/recipes/mountain_range/out/mountain_range_c/critique.json @@ -1,102 +1,62 @@ { "asset": "mountain_range_c", "reviewer": "agent", - "overall": 7.6, + "overall": 6.9, "scores": { "silhouette": { - "score": 8, - "reason": "2 \u00d7 2 km alpine tile with eroded ridges, a glacial U-valley, peaks above the snowline; reads as mountains from any angle.", - "fixes": [] + "score": 7, + "reason": "Two conical peaks with a valley read as mountains; the peaks are smooth cones; TOP shows a bright knife-edge seam along one ridge.", + "fixes": [ + "erode the cone flanks into ridges and couloirs", + "check the bright seam along the ridge crest in TOP (normal flip at the crest?)" + ] }, "proportions": { - "score": 8, - "reason": "Relief 950 m over a 1 720 m floor, tree line 2 000 m, snow line 2 480 m, plausible for the Alps.", - "fixes": [] + "score": 7, + "reason": "About 1 120 m relief on 2 × 2 km; peak flanks are evenly conical.", + "fixes": [ + "vary flank slope by aspect" + ] }, "materials": { - "score": 7, - "reason": "Ground strips chosen per triangle by slope and altitude: snow on gentle high ground, granite cliff on steep faces, gravel scree, grass meadows, moss on the forest floor; the cut sides show rock_strata beds. On 7.8 m cells the strips' V is squeezed per triangle, which draws fine banding on long slopes.", + "score": 6, + "reason": "Banding on the slopes; meadow green and a grey-blue patch on the left; little snow.", "fixes": [ - "blend strips across triangle borders (vertex weights) instead of hard per-triangle switches", - "finer UV scale on the skirt sides" + "blend strips across triangle borders", + "snow on the north faces of the two peaks" ] }, "detail": { - "score": 7, - "reason": "Hydraulic and thermal erosion channels, talus, ridges at 256\u00b2 cells; no trees or rocks scattered on the surface.", + "score": 6, + "reason": "Some ridges on the left peak; most surfaces are smooth.", "fixes": [ - "scatter tree impostors below the tree line and boulders on the scree" + "more erosion iterations; talus at slope feet" ] }, "wear": { - "score": 7, - "reason": "Erosion gullies, scree fans and snow only on gentle slopes tell the weather; no avalanche paths or streams.", + "score": 6, + "reason": "Erosion only.", "fixes": [ "a stream along the valley floor (water strip)" ] }, "lod": { "score": 8, - "reason": "Same grid at 256/128/64 cells; strips classified the same way at each LOD, so colour bands hold.", - "fixes": [] + "reason": "FAR LOD2 holds outline and colour.", + "fixes": [ + "keep the ridge crest at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS all three seeds: 133 120 / 33 792 / 8 704, one draw.", - "fixes": [] + "reason": "QA clean; 133k / 34k / 8.7k tris, one draw.", + "fixes": [ + "drop skirt triangles that are never seen from above" + ] } }, "renders": { - "hero.png": "c9852e5f4f7830b51c8813212dc06487b6b9103a41ac84eb67118966142d9da4", - "sheet.png": "c26b8e8ea83bcdc280f43af5ae797129fcb2ce451f333d4a967ec6f861655947", - "raking.png": "d1a9352e44159d8400046b108698d5882f181f9873cfc32aed413bf687e2ceae", - "raking_detail.png": "d9bcbbb66502cf704b03ce2869390cf6d0b1a74283850c7d14e82a228ffe1087", - "lods.png": "5ddfff6ce2678fd309b8c1b2228ac21bcc4accd4ebd6022ea0c5265238b538c6", - "silhouette.png": "2bb5862a992c1e93ffdf664d947d692663f95dadfbdb2646d9eec34a3568d93d", - "value.png": "846584afd1c85c38ef2759d09f6b212e2cfd8e5a856801635b3c44f7b91f12f2", - "detail.png": "3790232bf717f169c4e2b829ad0431075482ae0cde2c8bd63945974e160a886f" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.995 - }, - { - "view": "side", - "iou": 0.994 - }, - { - "view": "3/4", - "iou": 0.998 - } - ], - "focal": { - "peakPart": "terrain", - "peak": [ - 490, - 79 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "terrain", - "share": 1 - } - ], - "declared": [] - }, - "rest": { - "share": 0.701, - "spread": 0.692, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.089, - "straightShare": 0.658, - "taper": 0.562, - "shape": "stable" - } + "hero.png": "ccdda777804c48ab76fbe4c0dfeed0dea11d05e765c4b82e9bb2dd89653e1e90", + "sheet.png": "2c0ce1ed57c62192ddc8741998443e6b70767f2c3e8834b26713a3180521c054" } } diff --git a/recipes/oak/out/oak/critique.json b/recipes/oak/out/oak/critique.json index d3702ccb..abc595d9 100644 --- a/recipes/oak/out/oak/critique.json +++ b/recipes/oak/out/oak/critique.json @@ -1,118 +1,64 @@ { "asset": "oak", "reviewer": "agent", - "overall": 6.6, + "overall": 5.4, "scores": { "silhouette": { - "score": 6, - "reason": "Reads as a broad deciduous tree, but the `tree` kind's crown is lumpy and one-sided with large flat leaf cards; not yet the wide domed crown of an open-grown oak.", + "score": 5, + "reason": "Reads as a broadleaf tree in sheet FRONT/SIDE, but hero.png shows a lopsided bush of large flat leaf cards; not the wide domed crown of an open-grown oak.", "fixes": [ - "crown density and symmetry in src/kinds/vegetation (not owned by this recipe)", + "wider, domed crown: longer first-order limbs low on the bole", "more, smaller leaf clusters per twig" ] }, "proportions": { - "score": 7, - "reason": "22 m height, 22 m spread, 1.3 m DBH pass the measures; the bole looks too slender for an open-grown oak beside reference photos.", + "score": 6, + "reason": "22 m height/spread pass measures; the bole is slender and short for a 1.3 m DBH and the crown is taller than it is wide in hero.", "fixes": [ - "shorter, thicker bole and heavier first limbs (species preset)" + "thicker bole and heavier first limbs (species preset)" ] }, "materials": { - "score": 6, - "reason": "Bark now on the bark_oak strip (ridged furrows) with the kind's height gradient as a tint, moss on the north lower trunk, leaf-litter floor. Leaf cards are flat colour with no leaf texture (foliage cells unused).", + "score": 5, + "reason": "Leaf cards are flat colour with no leaf texture; sheet.png draws the bark/stem strip near-white while hero.png shows it brown, so the two images disagree on the trunk material.", "fixes": [ "leaf cards sampling the foliage cells", - "floor blended into the crown shadow" + "bark strip that renders the same in sheet and hero" ] }, "detail": { - "score": 6, - "reason": "Branch hierarchy, root flare, leaf clusters at LOD0. Leaf cards are large polygons that read as paper cut-outs up close.", + "score": 5, + "reason": "Branch hierarchy and root flare; leaf cards are large polygons that read as paper cut-outs.", "fixes": [ "finer leaf cards with alpha foliage textures" ] }, "wear": { - "score": 6, - "reason": "North-side moss on the lower trunk, leaf litter underfoot. No lichen, deadwood or bark damage.", + "score": 5, + "reason": "No moss, lichen or deadwood visible in the renders.", "fixes": [ - "a dead limb and a few bark scars", + "a dead limb and bark scars", "lichen patches on the limbs" ] }, "lod": { - "score": 7, - "reason": "LOD2 impostor keeps crown mass and trunk; floor drops to 6-gon.", + "score": 4, + "reason": "FAR LOD2 is a solid lumpy lathe blob that hides the trunk and changes the crown shape: a hard pop.", "fixes": [ - "impostor colour closer to the LOD0 crown mean" + "LOD2 crown from clustered leaf cards (like pine) fitted to the LOD0 silhouette", + "keep the trunk visible under the LOD2 crown" ] }, "technical": { "score": 8, - "reason": "QA PASS; kind budget plus \u2264 400 floor triangles; trunk DBH and total height pass the measures; one material. Now 13 950 / 3 004 / 184. Connections gate clean (was 247 findings): branches are closed tubes rooted on their parent's axis (buried in the wood), the trunk is capped, and the geometric leaf cards declare connect rest (a card meets its twig at one point by design).", - "fixes": [] + "reason": "QA clean; 14k / 3k / 159 tris, one material.", + "fixes": [ + "spend LOD2 above 159 tris on a card crown instead of the lathe blob" + ] } }, "renders": { - "hero.png": "04bb4054ed03278e154ed3a9c5df0c6c04d99feeedfb08e30276c794107d9690", - "sheet.png": "b20458fec2d86c61e7efbfa720e31d73ac7184e3ef86cd3f23c2b3e6d23bf229", - "raking.png": "ec2dfd3acaa65e08f9e752f5edd9e1b40291c3e59a684f1e1b4465375b010708", - "raking_detail.png": "502eb129e2409965d64fc2e75dfc7b9bf283118737b930d8fd6c3761bbe0a705", - "lods.png": "622a185f2883c614ecde7a6be278eab23f8b1f991914ceaa12555bd0bbae2911", - "silhouette.png": "cb73b56ebf0d716fd16a5465b5c5f7b7d0e74ddf98b364eb0b1dee61ee52b185", - "value.png": "ecd84ca8af2c2b345d132bdfd1b322a5f0476c79e089d96201e014e3f23d422a", - "detail.png": "e20a9cd94eef9017f6598013b64e4bd4537f0e766bf3366e0aaa37193ec92561" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.638 - }, - { - "view": "side", - "iou": 0.564 - }, - { - "view": "3/4", - "iou": 0.668 - } - ], - "focal": { - "peakPart": "leaves", - "peak": [ - 479, - 178 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "leaves", - "share": 0.895 - }, - { - "part": "branches", - "share": 0.103 - }, - { - "part": "ground", - "share": 0.001 - } - ], - "declared": [] - }, - "rest": { - "share": 0.195, - "spread": 0.753, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.166, - "straightShare": 0.37, - "taper": -0.233, - "shape": "friendly" - } + "hero.png": "81a06e3571f0d9cc4f2609305f2733ba11b2c1b0d67c604d8d3ea646c861645f", + "sheet.png": "6c03f659dafdb2050b3fd30e4a3a7a64afb3f8b2c1769348d51574d9723adaea" } } diff --git a/recipes/orc_head/out/orc_head/critique.json b/recipes/orc_head/out/orc_head/critique.json index bfecbdf1..585d3c4d 100644 --- a/recipes/orc_head/out/orc_head/critique.json +++ b/recipes/orc_head/out/orc_head/critique.json @@ -1,129 +1,66 @@ { "asset": "orc_head", "reviewer": "agent", - "overall": 6.9, + "overall": 6.6, "scores": { "silhouette": { "score": 8, - "reason": "Brute shape, spiked pauldrons, tusks and the knot read at a glance; LOD2 holds.", + "reason": "Brute head, tusks, topknot and spiked pauldrons read at a glance in every view.", "fixes": [ - "a clearer neck line under the jaw in the side view" + "a clearer neck line under the jaw in the SIDE view" ] }, "proportions": { "score": 7, - "reason": "Ears ×1.2 now read against the jaw; the knot is a tidy bun with two wraps and tail locks, though the tail still arcs high behind the bun.", + "reason": "Head, tusks and ears read against the jaw; the pauldrons sit outboard of the shoulders with a visible gap in FRONT, so the armour looks hung beside the body rather than worn.", "fixes": [ - "drop the tail arc 1 cm" + "move the pauldrons in 3–4 cm so they rest on the deltoid", + "drop the topknot tail arc 1 cm" ] }, "materials": { "score": 6, - "reason": "Painted steel with chipped lips and rust, leather strap, fur, bone spikes separate.", + "reason": "Painted steel with rivets and rust, leather baldric, fur collar and bone spikes separate; skin is one roughness; the baldric reads cream in sheet.png but brown leather in hero.png.", "fixes": [ - "roughness variation on the skin" + "roughness variation on the skin (oily forehead, matte cheeks)", + "one leather colour for the baldric across both renderers" ] }, "detail": { "score": 7, - "reason": "Scars, war paint, rivets, stitched baldric with a true rectangular buckle, fur clumps, wound bun.", + "reason": "Scars, war paint, rivets, buckle, fur clumps, wound bun and lamed pauldrons.", "fixes": [ - "finer hair strands on the tail locks" + "finer hair strands on the tail locks", + "stitching on the baldric edge" ] }, "wear": { "score": 6, - "reason": "Dents, chipped paint, rust drips, scars.", + "reason": "Dents, chipped paint and rust on the pauldrons, scars on the face.", "fixes": [ - "dirt on the fur and strap" + "dirt on the fur and strap", + "nicks on the tusks" ] }, "lod": { - "score": 7, - "reason": "LOD2 keeps the mass and armour; the bun simplifies per LOD (no thin turns that fold).", + "score": 6, + "reason": "FAR LOD2 keeps mass and armour but the pauldrons turn into flat hollow slabs and the skin breaks into blotchy faceted patches.", "fixes": [ - "keep the tail locks as two strands at LOD2" + "keep the pauldron dome as a closed 8-segment shell at LOD2", + "average the skin vertex colour per LOD2 face instead of sampling one corner" ] }, "technical": { - "score": 7, - "reason": "QA PASS honestly: 24 broad regions (up to 0.32 m) replaced by 574 regions ≤ 3 cm; buckle frame shear and two real seams fixed; no eye socket joints yet (lids fill the orbit).", + "score": 6, + "reason": "QA clean; 180k / 46k / 4.8k tris is far over a game bust budget; pauldron lames and fur clumps eat most of it.", "fixes": [ + "halve pauldron lame and fur clump segment counts at LOD0", "orbit bore + socket joints like the other busts" ] } }, "renders": { - "hero.png": "70b4acdbf07a3e943aedd6bcff2d9cc1ceae1fd562cba50a7fa33fb658a9c25a", - "sheet.png": "0fa6541f9b175e14e706101b0f7456096624c2236104c13196db5ff05bb8882a", - "raking.png": "978cd2fd63bb0e5edc28f7a8d8f5a26782b7fbf7f191efa703b4ed95ba70a438", - "raking_detail.png": "184fe9f12883c8aef5af697f8bf9b0baf301f01a04bc6769e7f38765bc218917", - "lods.png": "90084a257bee4630c553e45a570d9f0881de5ab65ba81688afb0e4618558caf3", - "silhouette.png": "424449bd979653668aaf1a26f9ddc1db7099fff8de4ed40b381be9087aff96c1", - "value.png": "2aa8849b0cb270c6ac146806d65533fc1a020b889b8e6069cc120fe6c283ca5c", - "detail.png": "8e8e998f31ef9fa5185737ba52cb4a0bbcda75898ef2f8e1bf4b201b6e9fe5a4" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.893 - }, - { - "view": "side", - "iou": 0.898 - }, - { - "view": "3/4", - "iou": 0.872 - } - ], - "focal": { - "peakPart": "head", - "peak": [ - 479, - 206 - ], - "focalShare": 0.567, - "focalArea": 0.301, - "top": [ - { - "part": "head", - "share": 0.567 - }, - { - "part": "hair", - "share": 0.101 - }, - { - "part": "pauldrons", - "share": 0.096 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.443, - "spread": 0.664, - "min": 0.25 - }, - "grounded": { - "rise": 0.246, - "bands": [ - 0.178, - 0.258, - 0.336, - 0.358, - 0.375 - ] - }, - "shape": { - "acuteShare": 0.256, - "straightShare": 0.39, - "taper": 0.642, - "shape": "friendly" - } + "hero.png": "90980d003f1ddf4d246d5af3eb1ef399f2f8dbfa8f1c5a1495dfa63ab9a0d9bd", + "sheet.png": "cba3c1ec909f0e25e5894933a8076307ba7c7fe59313c5f7e2656402baf464c0" } } diff --git a/recipes/palm/out/palm/critique.json b/recipes/palm/out/palm/critique.json index f3b01117..72ef35d4 100644 --- a/recipes/palm/out/palm/critique.json +++ b/recipes/palm/out/palm/critique.json @@ -1,110 +1,63 @@ { "asset": "palm", "reviewer": "agent", - "overall": 7.3, + "overall": 6.7, "scores": { "silhouette": { "score": 8, - "reason": "Coconut palm reads immediately: slender bowed ringed stem, crown of drooping pinnate fronds, coconuts under the frond bases.", - "fixes": [] + "reason": "Slender bowed stem and a crown of drooping pinnate fronds read as a coconut palm at once.", + "fixes": [ + "a few fronds drooping lower below the crown" + ] }, "proportions": { "score": 8, - "reason": "20 m height, 9 m crown, 0.35 m stem pass the measures; frond length looks right.", - "fixes": [] + "reason": "20 m height, 9 m crown and 0.35 m stem pass the measures.", + "fixes": [ + "slightly thicker base flare at the foot" + ] }, "materials": { - "score": 7, - "reason": "Stem on bark_oak tinted grey-brown (rings from the kind), fronds green, sand floor. Fronds lack midrib/leaflet colour variation.", + "score": 6, + "reason": "Fronds green with pinnate leaflets, sand floor; the stem is one dark brown with no ring scars visible; sheet.png draws the bark/stem strip near-white while hero.png shows it brown, so the two images disagree on the trunk material.", "fixes": [ - "dry brown lower fronds", - "a ringed-stem project material (palm scars)" + "ringed-stem material (leaf scars)", + "dry brown lower fronds" ] }, "detail": { - "score": 7, - "reason": "Pinnate leaflets, fruit, ringed stem.", + "score": 6, + "reason": "Pinnate leaflets read; coconuts and frond boots are not visible in hero.", "fixes": [ - "frond boots and fibres at the crown" + "coconut cluster under the frond bases large enough to read", + "frond boots and fibre at the crown" ] }, "wear": { - "score": 6, - "reason": "Sand underfoot; no dead hanging fronds or weathered lower stem.", + "score": 5, + "reason": "Sand underfoot; no dead hanging fronds.", "fixes": [ "two dead fronds hanging under the crown" ] }, "lod": { - "score": 7, - "reason": "LOD2 frond cards keep the star crown.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 replaces the lacy fronds with few solid dark cards: the crown thins and darkens.", + "fixes": [ + "more LOD2 frond cards with cut-out alpha", + "match LOD2 frond colour to LOD0 mean" + ] }, "technical": { "score": 8, - "reason": "QA PASS; kind budget plus \u2264 400 floor triangles; trunk DBH and total height pass the measures; one material. Now 5 572 / 2 332 / 215. Connections gate clean (tree kind: closed stems rooted on the parent axis, leaf cards declared connect rest).", - "fixes": [] + "reason": "QA clean; 5.6k / 2.4k / 205 tris, one material.", + "fixes": [ + "trim stem segments on the straight lower stem" + ] } }, "renders": { - "hero.png": "f2cd79c7f18184fc7f78c5d25fa1dc475a68e9b0d0b46a30ba25c67b7f4c141c", - "sheet.png": "1d9c4d4d0bce2850db780f9426ca26fd2478be5b58d36fbd16bd1ecb55fd37d2", - "raking.png": "cbc60dda875cb8dc273d258739f8019c8f2c4e0e439c0efa1911a94b6e0ab224", - "raking_detail.png": "1ed39a132d86e0a67c0bab84e90fdfd6e56802fe98c775d259706a2f4644b06a", - "lods.png": "424e8ddf7996753b6b90b26c538e9141ba8ad6d8f8b13f9e3331d942c0151438", - "silhouette.png": "2b491f0cf49b39b7738b84e5bb97b293afe2f5147522cad75f867cea7a34d47f", - "value.png": "1cee7bbaa4fd6296fa78e99dffb0ca9afe0cbd769a932cdc40ae23e07a254f57", - "detail.png": "af25665b27fd3a3b86fecdb3aca9ea05b7fa51716e6a2173eaab43f0d03fe144" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.665 - }, - { - "view": "side", - "iou": 0.58 - }, - { - "view": "3/4", - "iou": 0.65 - } - ], - "focal": { - "peakPart": "ground", - "peak": [ - 385, - 353 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "ground", - "share": 0.563 - }, - { - "part": "trunk", - "share": 0.347 - }, - { - "part": "roots", - "share": 0.089 - } - ], - "declared": [] - }, - "rest": { - "share": 0.474, - "spread": 0.906, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.696, - "straightShare": 0.488, - "taper": -0.694, - "shape": "aggressive" - } + "hero.png": "5e6303476f34a8369d54e1d11d0462201c1c4e1959ea8b2920ff018c2a12ac51", + "sheet.png": "55c285f987a32bdc292084a5192a9799564aa051abd66059acda095e4bfa279c" } } diff --git a/recipes/parts_catalogue/out/parts_catalogue/critique.json b/recipes/parts_catalogue/out/parts_catalogue/critique.json new file mode 100644 index 00000000..35fdec6f --- /dev/null +++ b/recipes/parts_catalogue/out/parts_catalogue/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "parts_catalogue", + "reviewer": "agent", + "overall": 6.7, + "scores": { + "silhouette": { + "score": 7, + "reason": "Each catalogue item (door, window, drum, cask, crate, chair, stool, bench, ladder, lamp, wheel, beams, pipe, hose) reads on its own; the board as a whole is a layout.", + "fixes": [ + "group items by family with spacing so no two outlines overlap in FRONT" + ] + }, + "proportions": { + "score": 8, + "reason": "Door, chair, drum and wheel sit at real sizes against the 0.5 m grid.", + "fixes": [ + "check the stool seat height (looks ~5 cm high for a bar stool)" + ] + }, + "materials": { + "score": 7, + "reason": "In hero.png wood, steel, rubber, glass and paint separate; in sheet.png wood (cask, crate, chairs) renders near-white.", + "fixes": [ + "make the wood strips render in the sheet renderer too", + "rubber roughness on the hose (reads glossy)" + ] + }, + "detail": { + "score": 6, + "reason": "Barrel hoops, crate battens, tyre tread, lamp shade and tools on the bench; labels unreadable.", + "fixes": [ + "bigger floor labels", + "bolt heads on the beams and pipe flanges" + ] + }, + "wear": { + "score": 5, + "reason": "Drum has holes and the floor is worn; the rest is clean.", + "fixes": [ + "rust on the drum and beams, scuffs on the crate" + ] + }, + "lod": { + "score": 7, + "reason": "FAR LOD2 keeps every item and colour; wood turns white like LOD0 in the sheet.", + "fixes": [ + "keep the hose and ladder rungs at LOD2" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean; 42k / 20k / 11k tris over 75 parts: LOD2 keeps a quarter of LOD0. SIDE -Y in sheet.png shows only four items.", + "fixes": [ + "cut LOD2 to ~4k (drop labels, tools on the bench)", + "check the SIDE -Y sheet view (door, window, ladder absent)" + ] + } + }, + "renders": { + "hero.png": "faf35820d289f377f586c4af85d03e9497f4d01124095d11bfbdf63d93303ff7", + "sheet.png": "a09359f35f457c44590ca1459b23a08622e2a4bff195c9fa1ea58502d30af213" + } +} diff --git a/recipes/pbr_showcase/out/pbr_showcase/critique.json b/recipes/pbr_showcase/out/pbr_showcase/critique.json new file mode 100644 index 00000000..00d1b7e3 --- /dev/null +++ b/recipes/pbr_showcase/out/pbr_showcase/critique.json @@ -0,0 +1,61 @@ +{ + "asset": "pbr_showcase", + "reviewer": "agent", + "overall": 7.6, + "scores": { + "silhouette": { + "score": 7, + "reason": "A board of labelled samples: spheres, a cylinder, a tube, a dome, a cube and a bowl; reads as a material test card.", + "fixes": [ + "raise the samples on taller plinths so they separate from the board in FRONT" + ] + }, + "proportions": { + "score": 8, + "reason": "Samples at a consistent ~20 cm scale, board 1.5 × 4.6 m.", + "fixes": [ + "add a 1.8 m figure or a ruler for scale" + ] + }, + "materials": { + "score": 8, + "reason": "Flake clearcoat red and blue, brushed metal, white gloss, glass, ice, milk and velvet read as themselves in hero.png.", + "fixes": [ + "an HDR environment band so the brushed metal and glass show anisotropy and refraction" + ] + }, + "detail": { + "score": 6, + "reason": "Labels too small to read; plain plinths.", + "fixes": [ + "larger labels", + "a bevel on each plinth" + ] + }, + "wear": { + "score": 8, + "reason": "A test card is clean by design.", + "fixes": [ + "a worn variant of each sample on a second row" + ] + }, + "lod": { + "score": 8, + "reason": "FAR LOD2 keeps every sample and its colour.", + "fixes": [ + "keep sphere segment count high enough that highlights stay round" + ] + }, + "technical": { + "score": 8, + "reason": "QA clean; 16k / 4.6k / 1.5k tris over 33 parts.", + "fixes": [ + "merge labels per word to cut LOD0 triangles" + ] + } + }, + "renders": { + "hero.png": "5c9068268fcbed0ef3f7b70b49058e26edc2ee7ee30a3fcce4a1ee7156b0f894", + "sheet.png": "57f45cadcd0dc2766f8bc212ec1f8af2805eefacd67946c05cef078fe6f7b3e6" + } +} diff --git a/recipes/pine/out/pine/critique.json b/recipes/pine/out/pine/critique.json index 783cb807..22281225 100644 --- a/recipes/pine/out/pine/critique.json +++ b/recipes/pine/out/pine/critique.json @@ -1,32 +1,34 @@ { "asset": "pine", "reviewer": "agent", - "overall": 6.6, + "overall": 6, "scores": { "silhouette": { - "score": 7, - "reason": "Tall clean, slightly sinuous bole with a high crown of clumped, layered needle masses and sky gaps between them (sheet FRONT/SIDE/3-4 REAR) — reads as a pine, no longer a bare frame. The crown is still taller than wide (9 m × ~11 m) and its top is rounded-irregular rather than the flat umbrella top of an old Scots pine.", + "score": 6, + "reason": "Tall clear bole with a high clumped crown; the crown is narrow, lopsided and starburst-spiky, closer to a holly than the flat umbrella of an old Scots pine.", "fixes": [ - "flatter, wider crown top for the old-tree umbrella (needs a crown shape whose branch length peaks near the top; the kind's ShapeRatio has none)" + "flatter, wider crown top (branch length peaking near the top)", + "more even crown spread around the bole" ] }, "proportions": { "score": 7, - "reason": "25 m height, 9 m spread, 0.6 m DBH pass the measures; natural crown is now grown ~9 m across so the fit no longer squashes it ×0.52 horizontally. Crown base at ~55% of height, branch angles 80° low → 50° high per the README sources.", + "reason": "25 m height, 9 m spread and 0.6 m DBH pass measures; crown depth at the deep end for a mature tree.", "fixes": [ - "crown depth ~45% of height is at the deep end for a mature open-grown tree" + "shallower crown (~35 % of height)" ] }, "materials": { - "score": 6, - "reason": "Bark on bark_pine with the grey-brown → orange gradient and north moss; the grey seam streak down the trunk is gone. Needles are flat-shaded crossed brush cards on the plain strip: dark blue-green, with sheen, but no needle texture.", + "score": 5, + "reason": "Needle tufts are flat dark cards with no needle texture; sheet.png draws the bark/stem strip near-white while hero.png shows it brown, so the two images disagree on the trunk material.", "fixes": [ - "needle foliage texture / alpha cards for the tufts" + "needle foliage texture / alpha cards", + "bark strip that renders the same in sheet and hero" ] }, "detail": { "score": 6, - "reason": "404 needle tufts (two crossed brushes each) in clumps at the ends of 100 side shoots on 18 whorl branches; branches visible through the gaps (raking_detail). Up close the tufts read as spiky starbursts, not fine needles. No cones, no dead lower stubs, no bark plates in geometry.", + "reason": "Needle tufts clumped on shoots; no cones, no dead lower stubs.", "fixes": [ "cones on outer shoots", "dead branch stubs along the clear bole" @@ -34,88 +36,30 @@ }, "wear": { "score": 5, - "reason": "Moss on the north side of the foot and a needle-litter floor; otherwise clean.", + "reason": "Litter floor; otherwise clean.", "fixes": [ - "dead stubs and resin scars on the bole", - "a broken/dead branch in the crown" + "resin scars and dead stubs on the bole", + "a dead branch in the crown" ] }, "lod": { - "score": 7, - "reason": "LOD2 is 20 needle pads clustered from LOD0's own tufts and silhouette-fitted from eight views: IoU front 0.879, side 0.865, 3/4 0.891 (was 0.36/0.40/0.42 with the lathe blob), and the measured crown colour matches LOD0 (rgb 0.154/0.182/0.160 vs 0.154/0.178/0.160). Close up (lods.png) the pads read as pointed cones, a pagoda more than pads, and one fitted pad is a wide flat sheet. LOD1 now keeps LOD0's tufts (one per shoot, grown) instead of sparse light cluster cards (crown luminance 0.173 vs 0.171).", + "score": 5, + "reason": "FAR LOD2 is a cluster of pointed shard-like planes, a pagoda rather than needle pads; colour holds but shape pops.", "fixes": [ - "blunter pad tops (a two-ring lens) if the 316-triangle LOD2 budget allows", - "clamp the plan/height aspect a pad may be fitted to" + "blunter pad tops (a two-ring lens) at LOD2", + "clamp the plan/height aspect a fitted pad may take" ] }, "technical": { "score": 8, - "reason": "QA PASS; 23 652 / 5 740 / 311 triangles against 25 400 / 6 200 / 316 (≤ 8% headroom); one material, two draws; deterministic (LOD2 fit is pure arithmetic). LOD2 compile takes ~1.5 s for the silhouette fit.", + "reason": "QA clean; 24k / 5.8k / 306 tris, one material.", "fixes": [ - "cache or simplify the LOD2 silhouette fit (~1.5 s per compile)", - "spend the LOD0 headroom on cones and dead stubs rather than more tufts" + "cache or simplify the LOD2 silhouette fit" ] } }, "renders": { - "hero.png": "c0f4546dcc9ff1834028311f882255bf194d2b9d88758201bef6a9bfb54f369a", - "sheet.png": "965ee7e61238c3b366722c7b85cb071a613cd1bbc30cc83c78afe6a5e0ff29be", - "raking.png": "4cf0d9d95d8c29979adce53671842683129b8c5ab8ceed7997149b7bfd4c445e", - "raking_detail.png": "958c93f564adc16cce18929621aba81b99f74a5e87390f0a8213c04c6b3dd381", - "lods.png": "7db828e39dfc2f0c62d2f095e26d9b305f53fa9bcbae5a83715742774196b9ff", - "silhouette.png": "a3654c1d5d2583baa4dab839ad20847c8595d00b49df980c5550f091d6034e2a", - "value.png": "9c295bb93f7b79976a735f444ea0240d3fcc6a3f0de975f7f6366676ae64e784", - "detail.png": "a8aaff136fb8cf63c495c27e1c226b07a1ab81e784ac1c11db896a2c4f313df4" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.879 - }, - { - "view": "side", - "iou": 0.865 - }, - { - "view": "3/4", - "iou": 0.891 - } - ], - "focal": { - "peakPart": "leaves", - "peak": [ - 399, - 95 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "leaves", - "share": 0.664 - }, - { - "part": "ground", - "share": 0.125 - }, - { - "part": "roots", - "share": 0.114 - } - ], - "declared": [] - }, - "rest": { - "share": 0.139, - "spread": 0.612, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.072, - "straightShare": 0.572, - "taper": -0.824, - "shape": "stable" - } + "hero.png": "959bc54c6470888b36ab3849c6c33cf73a382041dd133910891ec713b33dbf29", + "sheet.png": "f16f86ba5d2a73dd19f0e3c550751a654be0f1f1e809daf3adeda9a16c22d73c" } } diff --git a/recipes/ranger/out/ranger/critique.json b/recipes/ranger/out/ranger/critique.json new file mode 100644 index 00000000..18240e0a --- /dev/null +++ b/recipes/ranger/out/ranger/critique.json @@ -0,0 +1,62 @@ +{ + "asset": "ranger", + "reviewer": "agent", + "overall": 7.6, + "scores": { + "silhouette": { + "score": 8, + "reason": "Hip-length jerkin with shoulder capes, baldric, belt with knife, knee boots in an A-pose: a ranger reads from every view.", + "fixes": [ + "a hood or cloak edge for a stronger back silhouette" + ] + }, + "proportions": { + "score": 8, + "reason": "1.78 m with sourced segment heights; head, hands and feet look right.", + "fixes": [ + "slightly larger hands (they read small in FRONT)" + ] + }, + "materials": { + "score": 8, + "reason": "Textured set: leather jerkin, wool trousers, leather boots and belt, metal buckles, skin and hair read as themselves.", + "fixes": [ + "higher roughness contrast between wool trousers and leather boots" + ] + }, + "detail": { + "score": 7, + "reason": "Studs, buckles, baldric hardware, collar and bracer laces; hands are simple.", + "fixes": [ + "finger separation and knuckles at LOD0", + "visible stitching at the jerkin seams" + ] + }, + "wear": { + "score": 6, + "reason": "Texture has some leather scuffing; boots and hem are clean.", + "fixes": [ + "mud on the boots and trouser hem", + "scuffed leather on the belt" + ] + }, + "lod": { + "score": 8, + "reason": "FAR LOD2 keeps outline, colour and face read.", + "fixes": [ + "keep the knife hilt at LOD2" + ] + }, + "technical": { + "score": 8, + "reason": "QA clean; 39k / 12k / 3.9k tris, one textured part.", + "fixes": [ + "verify the texture set size against budget.textureSets" + ] + } + }, + "renders": { + "hero.png": "998f9a70746d4b00687fa50979492360ead2d1296b4faf78a7912567aa55f991", + "sheet.png": "28423d5400a53fcbe2fa880dd9ad41430676cc16316c79c0e21ed776fa192ad2" + } +} diff --git a/recipes/real_valley/out/real_valley/critique.json b/recipes/real_valley/out/real_valley/critique.json index 43afb94a..cabd0b68 100644 --- a/recipes/real_valley/out/real_valley/critique.json +++ b/recipes/real_valley/out/real_valley/critique.json @@ -1,25 +1,25 @@ { "asset": "real_valley", "reviewer": "agent", - "overall": 7.4, + "overall": 7.3, "scores": { "silhouette": { "score": 8, - "reason": "Measured USGS 3DEP relief: Half Dome central and highest, sheer north-rim walls over a flat glacial floor; reads as Yosemite's east end from the Glacier Point direction.", + "reason": "Measured USGS relief: Half Dome and the north-rim walls over the flat floor read as Yosemite in the sheet views.", "fixes": [ - "z14 tiles to sharpen Half Dome's NW face lip (stepped at 15.6 m spacing)" + "frame the hero from the Glacier Point side so Half Dome is in the outline" ] }, "proportions": { "score": 9, - "reason": "True scale 4 \u00d7 4 km; Half Dome summit 1 524.8 m above datum vs 1 525 m from Wikidata (\u22120.01 %); Glacier Point within 0.9 % of Wikidata.", + "reason": "True scale 4 × 4 km; summit heights match Wikidata within 1 %.", "fixes": [ - "a true-relief view from Glacier Point's own position (an interior camera) to compare the photo silhouette 1:1" + "a true-relief view from Glacier Point to compare 1:1" ] }, "materials": { - "score": 6, - "reason": "Strips per triangle by slope and altitude (cliff, gravel, moss); granite domes' 30\u201345\u00b0 bare backs read partly as forest; LOD2 shows diagonal banding in the moss/grass strips.", + "score": 5, + "reason": "Hero.png shows strong diagonal stripe banding on every slope and granite reads olive; sheet.png renders the terrain flat grey.", "fixes": [ "land cover from an open dataset instead of slope-only colouring", "blend strips across triangle borders" @@ -27,85 +27,36 @@ }, "detail": { "score": 7, - "reason": "Survey detail only (talus cones, gullies, benches at 15.6 m); no trees, boulders or river.", + "reason": "Survey detail only (talus, gullies, benches); no trees, boulders or river.", "fixes": [ - "scatter tree impostors on the forest floor", - "carve the Merced from an OSM centreline" + "tree impostors on the forest floor", + "the Merced carved from an OSM centreline" ] }, "wear": { "score": 6, - "reason": "No weathering beyond what the terrain shape carries; no water streaks on the granite.", + "reason": "Terrain shape carries the weathering; no water streaks on the granite.", "fixes": [ - "dark water streaks down the cliff strip below drainage lines" + "dark water streaks down the cliffs below drainage lines" ] }, "lod": { "score": 8, - "reason": "256/128/64 cells from one grid; LOD2 silhouette IoU \u2265 0.99; small dark slivers on LOD2 ridgelines.", + "reason": "FAR LOD2 keeps silhouette and relief.", "fixes": [ - "drop or seat the dark slivers on LOD2 ridgelines (strip seams at coarse cells)" + "seat the dark slivers on LOD2 ridgelines" ] }, "technical": { "score": 8, - "reason": "One part, one draw call, 133k/34k/8.7k triangles, byte-deterministic, DEM hash verified on decode, credit in the GLB copyright.", + "reason": "QA clean; 133k / 34k / 8.7k tris, one draw, DEM hash verified.", "fixes": [ - "a 512-cell z14 variant for hero use with the 256 grid kept for games" + "a 512-cell z14 variant for hero use" ] } }, "renders": { - "hero.png": "74f40bd66620ea3c1aeff371c43b31ee3e7bffb2f17453826a1c6270b9a6d5b0", - "sheet.png": "9eea544b1abe4f5492886fb061aa45aa22172507ac82609496438d6c61f32ac2", - "raking.png": "e7419868a9d7797416f99d918f1c49729a1d358048aceaf8a369449efe7eec7b", - "raking_detail.png": "4f0c96d41d524b078a6e6541077f8d1b0e7e030022631c6a8d5cf9c1a13f3581", - "lods.png": "5be3c978798eb4a764e90b6d3291e8d8e33e259e4959a855a595c47db3806b00", - "silhouette.png": "2bb2962172afd580d1e574ad6ae0bd7f2349b797978736b480455e5994d28c1c", - "value.png": "18aff109d39bf43170f300c2099f3d89b10f82ba1d5fcd796f42fc0b4e360653", - "detail.png": "4ad4a4c38264c65128b3ae3d21ad069727dfaab84070c88aa389fe687e733cd6" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.998 - }, - { - "view": "side", - "iou": 0.997 - }, - { - "view": "3/4", - "iou": 0.999 - } - ], - "focal": { - "peakPart": "terrain", - "peak": [ - 309, - 241 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "terrain", - "share": 1 - } - ], - "declared": [] - }, - "rest": { - "share": 0.746, - "spread": 0.616, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.067, - "straightShare": 0.719, - "taper": 0.647, - "shape": "stable" - } + "hero.png": "44aa8b9bae95952c8db0eb7126b58ce2ff4eed9ae883a118e1112f8bf2eb0d31", + "sheet.png": "b77de0d3c7dd8658772996d1ce3f8964577f0f3264dafb315d51af9f2c7181ec" } -} \ No newline at end of file +} diff --git a/recipes/roadster_v10/out/roadster_v10/critique.json b/recipes/roadster_v10/out/roadster_v10/critique.json new file mode 100644 index 00000000..20ded0f6 --- /dev/null +++ b/recipes/roadster_v10/out/roadster_v10/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "roadster_v10", + "reviewer": "agent", + "overall": 6.3, + "scores": { + "silhouette": { + "score": 8, + "reason": "Long bonnet, cab-rearward roadster with roll hoops, raked screen and side exhaust reads as a front-engined V10 roadster.", + "fixes": [ + "a more pronounced rear haunch over the rear wheel" + ] + }, + "proportions": { + "score": 7, + "reason": "4.5 m long, 1.3 m high; wheels and overhangs look right; the front is tall and blunt in FRONT.", + "fixes": [ + "lower the nose 3–4 cm and sharpen the front overhang plan shape" + ] + }, + "materials": { + "score": 7, + "reason": "Red paint, black grille mesh, alloy wheels, rubber, glass and chrome exhaust separate; the paint reads satin without clearcoat reflections; the windscreen renders opaque light grey in sheet.png.", + "fixes": [ + "clearcoat finish with sharp reflections on the body paint", + "consistent transparent glass in both renderers" + ] + }, + "detail": { + "score": 7, + "reason": "Bonnet vents, side gills, grille mesh, headlamp internals, mirrors, calipers, side exhaust, interior.", + "fixes": [ + "shut-lines for doors and bonnet", + "badges and fuel filler" + ] + }, + "wear": { + "score": 4, + "reason": "Showroom-clean apart from a light dust speckle.", + "fixes": [ + "brake dust on the wheels and road grime on the sills" + ] + }, + "lod": { + "score": 6, + "reason": "FAR LOD2 turns the tyres light grey with the wheels: the black rubber disappears and the wheels read as white discs.", + "fixes": [ + "keep a black tyre ring separate from the rim at LOD2" + ] + }, + "technical": { + "score": 5, + "reason": "QA clean but 909k / 78k / 5.5k tris: LOD0 is roughly 5–10× a hero car budget and drops 12× to LOD1.", + "fixes": [ + "gate wheel spokes, grille mesh and interior ring counts by ctx.lod to bring LOD0 under ~150k", + "merge static body draws (91 parts)" + ] + } + }, + "renders": { + "hero.png": "985611e643917323773d1fcfef8505365978152a852d9f4661ee8df3964d7a11", + "sheet.png": "6159d4f047c5988a09884ff18864db92442bffe98bb83295c710c4138c5ef6d7" + } +} diff --git a/recipes/roadster_v10/out/roadster_v10_open/critique.json b/recipes/roadster_v10/out/roadster_v10_open/critique.json new file mode 100644 index 00000000..46e8ebff --- /dev/null +++ b/recipes/roadster_v10/out/roadster_v10_open/critique.json @@ -0,0 +1,62 @@ +{ + "asset": "roadster_v10_open", + "reviewer": "agent", + "overall": 6.1, + "scores": { + "silhouette": { + "score": 7, + "reason": "Doors, bonnet and boot open: reads as the roadster with everything open; the raised bonnet is a huge slab that hides the front end.", + "fixes": [ + "hinge the bonnet at the rear (or check the front-hinge angle) so the nose stays visible" + ] + }, + "proportions": { + "score": 7, + "reason": "Same body as the closed car; the open bonnet looks larger than its opening.", + "fixes": [ + "fit the bonnet panel to the opening outline" + ] + }, + "materials": { + "score": 7, + "reason": "Same paint, glass and mesh; the open bonnet surface is visibly lumpy/warped in hero.png.", + "fixes": [ + "smooth the bonnet panel surface (warped shading)", + "clearcoat finish on the paint" + ] + }, + "detail": { + "score": 7, + "reason": "Engine bay with V10 heads, intake, reservoirs; door cards and hinges.", + "fixes": [ + "door hinges and latches at the shut faces", + "underside of the bonnet with bracing" + ] + }, + "wear": { + "score": 4, + "reason": "Showroom clean; the engine bay has no grime.", + "fixes": [ + "heat discolouration on the exhaust headers and grime in the bay" + ] + }, + "lod": { + "score": 6, + "reason": "FAR LOD2 keeps the open panels; tyres turn light grey with the rims.", + "fixes": [ + "keep a black tyre ring separate from the rim at LOD2" + ] + }, + "technical": { + "score": 5, + "reason": "QA clean but 909k / 78k / 6.3k tris over 91 parts, same over-budget LOD0 as the closed car.", + "fixes": [ + "gate engine-bay and grille mesh density by ctx.lod to bring LOD0 under ~150k" + ] + } + }, + "renders": { + "hero.png": "bf3cc03da4cd25289d260fe726656f3f32ff381595748e571de373b6df683a50", + "sheet.png": "9b6fd40c87c02e1b7b3e9c6ed4c30cf1dc05ff3f3fb7a63ca8c99b11070f9ad7" + } +} diff --git a/recipes/rts_barracks/out/rts_barracks/critique.json b/recipes/rts_barracks/out/rts_barracks/critique.json index 3cd37554..102d8860 100644 --- a/recipes/rts_barracks/out/rts_barracks/critique.json +++ b/recipes/rts_barracks/out/rts_barracks/critique.json @@ -1,105 +1,65 @@ { "asset": "rts_barracks", "reviewer": "agent", - "overall": 7.3, + "overall": 6.6, "scores": { "silhouette": { "score": 7, - "reason": "Gable + two coursed drum towers + porch read as one family across all nine states.", - "fixes": [] + "reason": "Gable hall, two coursed drum towers with conical red caps and finials, a porch and two banners: reads as a stylised RTS barracks from every sheet view. The porch roof is a thin slab that reads as a tilted plank in the hero, and the gate under it barely shows, so the front loses its focal entrance.", + "fixes": [ + "give the porch a real pitched roof (two slopes + ridge) and deepen the gate recess in roof.ts porch()/gate() so the entrance reads in black" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "15.9 x 18.5 x 11.1 m, gate 3.5 m beside 1.8 m units; storey and door scale per the house playbook. Towers are tall versus the hall and the windows are narrow amber slits, which reads stylised but slightly toy-like.", + "fixes": [ + "widen the window openings to ~0.8 m with a stone sill and head (walls.ts) so the bay rhythm matches a real hall" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 6, + "reason": "Dressed-stone blocks read as stone, lime-plaster gables in an oak frame read, red team slate reads more like painted sheet metal panels (uniform value, no slate edge variation). In sheet.png the oak porch posts render pale grey-white, not brown.", "fixes": [ - "window reveals / sills as their own stone" + "per-slate value jitter and overlapping thickness on roofSlopes() so the roof reads as slate, not sheet panels", + "check the oak role on porch posts/timber: they read pale grey in sheet.png while brown in hero.png" ] }, "detail": { - "score": 7, - "reason": "Running-bond blocks, pilasters, timber-framed gables, oculus, beacons, gate with straps, banners with sleeves and sewn chevrons.", - "fixes": [] + "score": 6, + "reason": "Running-bond blocks, pilasters, timber gable frame, oculus, finials and banners exist; at hero distance the walls are flat between pilasters, windows have no frames or sills, and the towers have no openings or arrow slits.", + "fixes": [ + "window frames, sills and lintels (walls.ts)", + "arrow slits and a door on each tower (roof.ts tower())", + "a bracketed porch lamp and steps with nosings at the gate" + ] }, "wear": { - "score": 7, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy.", - "fixes": [] + "score": 5, + "reason": "weather() is applied, but at hero distance the stone and plaster read clean; no visible moss at the wall foot, no streaks below windows, roof uniform red.", + "fixes": [ + "raise weather() streak and damp coverage under eaves and sills so it shows at 3/4 hero distance", + "fade and dust the roof slates toward the ridge, darken the drip line along the eaves" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps hall, towers, roof and banners in the same colour blocks; the stone coursing collapses to flat beige boxes per bay and the towers become coarse octagons, so the stone read is lost at distance.", + "fixes": [ + "bake the block coursing as value stripes in vertex colour at LOD2 (walls.ts stack())", + "12-sided towers at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 41.6k / 8.25k / 896, draws 3 / 3 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "5ff55e0548ffe406de41ec7f89c4bb1fc0166c9dce26503032c7eb757845fb40", - "sheet.png": "fb378c53a5e9b8039e3eff48faa337b12e080bade63ddbac77f0cf4d0cbcaa42", - "raking.png": "63e12a7b7c23481c6081ee781ec9c510966dd6d80ea0d11f9ad53a8d41687708", - "raking_detail.png": "89a68f8857c1fb74230b421e178deb2d73405319df3bbbff60f0f545c65f7f6d", - "lods.png": "58d723388197721f84de9869c23ad38852d801107cc70adea0048869447af725", - "silhouette.png": "08bfb20a567c8e91452498eb4d8b9955709e5a4187e5e4e48c264f5afd303350", - "value.png": "701b94f138187673d07ee642635b4c7ddf9357b1097eb375efc67851325d309c", - "detail.png": "4767090d2d5e5ea23539c52290d56c2c96aed6eea06c6d674e7c520b43750fa3" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.983 - }, - { - "view": "side", - "iou": 0.986 - }, - { - "view": "3/4", - "iou": 0.982 - } - ], - "focal": { - "peakPart": "walls", - "peak": [ - 264, - 278 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.425 - }, - { - "part": "pilasters", - "share": 0.407 - }, - { - "part": "porch", - "share": 0.089 - } - ], - "declared": [] - }, - "rest": { - "share": 0.377, - "spread": 0.743, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.29, - "straightShare": 0.708, - "taper": 0.847, - "shape": "stable" - } + "hero.png": "a712f93aadbf6aed84e8456bcd80cd6420deb6a95e40acbd36311d04899d24d8", + "sheet.png": "479f6478bc372278e0a0fa111fcbbfee6f20a3e5ea37f44ffd64030849f8e0a6" } } diff --git a/recipes/rts_barracks/out/rts_barracks_damaged/critique.json b/recipes/rts_barracks/out/rts_barracks_damaged/critique.json index 8395b995..508262eb 100644 --- a/recipes/rts_barracks/out/rts_barracks_damaged/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_damaged/critique.json @@ -1,105 +1,63 @@ { "asset": "rts_barracks_damaged", "reviewer": "agent", - "overall": 7.3, + "overall": 6.6, "scores": { "silhouette": { "score": 7, - "reason": "Holes, missing roof strips over exposed rafters, broken tower.", - "fixes": [] + "reason": "Same hall with a roof hole showing rafters, one tower sheared to a flat stump, notched walls and a debris heap at the porch: reads as damaged from every view.", + "fixes": [ + "jagged break line on the sheared tower (stepped blocks instead of a flat cut) so the damage reads in silhouette" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "15.9 x 18.3 x 11.1 m, same plan as the complete hall; the roof hole and missing blocks sit within plausible bays.", + "fixes": [ + "widen the window openings to ~0.8 m with sills (walls.ts) as on the complete hall" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 6, + "reason": "Stone, plaster and oak read as on the complete hall; the exposed rafters read as timber. Roof slate still reads as sheet panels, and the torn edges of the roof are clean straight cuts.", "fixes": [ - "window reveals / sills as their own stone" + "broken, offset slates along the edge of the roof hole (roofSlopes missing bays)", + "per-slate value jitter on the roof" ] }, "detail": { - "score": 7, - "reason": "Holes from the cell plan, exposed rafters, debris heap with charred beams, torn banner; smoke x4 + fire emitters.", - "fixes": [] + "score": 6, + "reason": "Roof hole with rafters, notched wall courses, a debris heap and a sheared tower; damage is rectangular (whole blocks missing) and there are no splintered rafters or hanging slates.", + "fixes": [ + "splintered rafter ends and 2-3 hanging slates at the roof hole", + "loose blocks on the wall tops next to each notch" + ] }, "wear": { - "score": 7, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Scorch zones round the holes, rubble dust (50% of up-faces).", - "fixes": [] + "score": 5, + "reason": "Notches and the debris heap tell the story, but there is no scorch, soot or dust around the damage and the walls are as clean as the undamaged hall.", + "fixes": [ + "grime() scorch above the roof hole and the sheared tower, soot streaks down the wall from the notches", + "dust spray around the debris heap" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps the roof hole, the stump tower and the debris heap; coursing becomes flat beige boxes as on the complete hall.", + "fixes": [ + "bake the block coursing as value stripes in vertex colour at LOD2 (walls.ts stack())" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 38.4k / 8.16k / 866, draws 3 / 3 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "551ee2b1230aeacf14d44888907d63210ea499e7bcd20ad647e3b59816b614aa", - "sheet.png": "8fa9d2e426955f723252e07cae5b2a7a76e8e40834110fefdd1f5917b8a2fef6", - "raking.png": "16976bba287654ff18df9797094e5ab2f8ffcb2ca8d615a5509d62559964e38c", - "raking_detail.png": "4abcb05a891b246c54e90bcc5302dbbf7f7a1033e8e215fc957db299c032ace6", - "lods.png": "f26cc1b81567fa0122ad939d9f510bac453251c38c35facd2e5912e9d21996e4", - "silhouette.png": "f35ad718e3f16ce3f790f3c571079440a2d702101630ecd28c480fdc6c1d9a06", - "value.png": "ed4c4a23a217f14c33d6fea278b86bbf0e4f67f6ff173fa5da567a4aa6b517a3", - "detail.png": "c56847b9e7661a78b6ad329cb6477a486c1aa70f17cb0c93c45a94a9626bc60a" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.985 - }, - { - "view": "side", - "iou": 0.944 - }, - { - "view": "3/4", - "iou": 0.981 - } - ], - "focal": { - "peakPart": "debris", - "peak": [ - 437, - 357 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.23 - }, - { - "part": "pilasters", - "share": 0.213 - }, - { - "part": "debris", - "share": 0.137 - } - ], - "declared": [] - }, - "rest": { - "share": 0.346, - "spread": 0.737, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.231, - "straightShare": 0.732, - "taper": 0.864, - "shape": "stable" - } + "hero.png": "7e40ae545310aa93f75ebaf4c876858b1a86eed82ae419e35f23f6e10087d5a7", + "sheet.png": "764a82eab51d0c0aa15f62db624e710dfac082d2081918b1a0b9fcfa19568bea" } } diff --git a/recipes/rts_barracks/out/rts_barracks_foundation/critique.json b/recipes/rts_barracks/out/rts_barracks_foundation/critique.json index f0df97e9..07e820e2 100644 --- a/recipes/rts_barracks/out/rts_barracks_foundation/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_foundation/critique.json @@ -1,105 +1,64 @@ { "asset": "rts_barracks_foundation", "reviewer": "agent", - "overall": 6.9, + "overall": 5.9, "scores": { "silhouette": { - "score": 6, - "reason": "Plinth, first course, team-tarped pallets; low and flat by design.", - "fixes": [] + "score": 5, + "reason": "Flat slab with a team triangle inlay, two tower footprint discs, low block courses and three red-topped piles: reads as a build site, but the piles read as small tables with red tops rather than tarped pallets, and nothing stands up except one banner.", + "fixes": [ + "drape the tarps over the block piles (sagging edges hanging down the sides) in site.ts materials()", + "add corner setting-out stakes with string lines or a wheelbarrow so the site has verticals" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "Slab and tower discs match the finished hall footprint; block piles at 1.4 m are plausible pallets.", + "fixes": [ + "stack the timber to a believable 0.7-1.0 m height instead of two thin courses" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 5, + "reason": "Slab reads as dark concrete, blocks as stone; the timber stack behind the slab renders pale blue-lilac in hero.png instead of oak, and the tarps read as rigid red slabs, not canvas.", "fixes": [ - "window reveals / sills as their own stone" + "fix the timber stack colour in site.ts materials(): it renders pale blue in hero.png, not WOOD/oak", + "canvas tarps with folds and draped edges instead of 6 cm boxes" ] }, "detail": { - "score": 6, - "reason": "Pallets of blocks on pine pallets under team tarps, timber stack, survey banner, floor chevron.", - "fixes": [] + "score": 5, + "reason": "Inlay, footing courses, three pallets, a timber stack and a banner. No tools, mortar tubs, ropes or setting-out lines; the slab is otherwise empty.", + "fixes": [ + "mortar tub, shovel and a wheelbarrow beside one pile", + "string lines between corner stakes along the wall lines" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy.", - "fixes": [] + "score": 4, + "reason": "The slab and blocks are clean; no mud, footprints, spilled mortar or dust around the piles.", + "fixes": [ + "grime() mud and mortar splash around the pallets and along the slab edge", + "dust on the slab top" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps the slab, inlay, tower discs, piles and banner in the same colours; piles become single blocks.", + "fixes": [ + "keep two blocks per pile at LOD2 so the piles keep their stepped outline" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 5.03k / 1.77k / 500, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "df03a65a3d7c7bd4ac1c3115eb3fb9ce44c3939761ce20b36c6d1317aa8cef2c", - "sheet.png": "81ddb5ae401da6dca46de669b95349dd9ceff1df528b2fb9efdac25e18006e38", - "raking.png": "d391e9a1cef260079f14dcc26640114f6df0d7183658dc6e4ed22a15276d50b2", - "raking_detail.png": "8d8cbb11b41f9dc6935f5aa2d70bc3df8364e5b36c74411392698a81fe483d50", - "lods.png": "0666ace82ffa60b7645b51f002ed5a664356710b1066ac261b108002c9c7c2ef", - "silhouette.png": "d7d777e1507dd7ee0317a7bf704786bb9f94d43a15b556a01da1c5241c913362", - "value.png": "79dca3cbc0435732aa768bf6b20414e555c4e8f444550e4000a8a9d2a509b67c", - "detail.png": "89704556ba198307c43d76cc7266dda1fd5b47e0b886ae460a9e110ddeb4b431" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.991 - }, - { - "view": "side", - "iou": 0.924 - }, - { - "view": "3/4", - "iou": 0.956 - } - ], - "focal": { - "peakPart": "tower_r", - "peak": [ - 117, - 186 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.38 - }, - { - "part": "tower_r", - "share": 0.216 - }, - { - "part": "plinth", - "share": 0.182 - } - ], - "declared": [] - }, - "rest": { - "share": 0.685, - "spread": 0.75, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.088, - "straightShare": 0.748, - "taper": 0.705, - "shape": "stable" - } + "hero.png": "65ba603374226d24fcf742f1a830336894434f50864600084086c280e6c2402d", + "sheet.png": "480a421a46a0ce72847f9334aabe2227f2c8d7893a42c7332ee5f174c2773b27" } } diff --git a/recipes/rts_barracks/out/rts_barracks_foundation_damaged/critique.json b/recipes/rts_barracks/out/rts_barracks_foundation_damaged/critique.json index 7d9ba6de..52ccbfd8 100644 --- a/recipes/rts_barracks/out/rts_barracks_foundation_damaged/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_foundation_damaged/critique.json @@ -1,105 +1,62 @@ { "asset": "rts_barracks_foundation_damaged", "reviewer": "agent", - "overall": 6.9, + "overall": 5.9, "scores": { "silhouette": { - "score": 6, - "reason": "Raided foundation.", - "fixes": [] + "score": 5, + "reason": "The foundation slab with one pallet knocked over into a small debris heap; otherwise identical to the foundation, and the damage barely reads from the 3/4 hero.", + "fixes": [ + "knock a corner off the slab and tip a pallet over the edge so the damage reads in outline" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "Same slab and footprint as the foundation; debris block sizes match the pallets.", + "fixes": [ + "stack the timber to a believable 0.7-1.0 m height instead of two thin courses" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 5, + "reason": "As on the foundation: the timber stack renders pale blue-lilac and the tarps read as rigid red slabs.", "fixes": [ - "window reveals / sills as their own stone" + "fix the timber stack colour in site.ts materials(): it renders pale blue in hero.png, not WOOD/oak", + "canvas tarps with folds and draped edges" ] }, "detail": { - "score": 6, - "reason": "Blast scorch on the plinth, first course partly knocked off, debris, smoke.", - "fixes": [] + "score": 5, + "reason": "Foundation detail plus a small debris pile by the front pallet; no cracked slab, no scattered blocks.", + "fixes": [ + "crack lines and a broken corner on the slab", + "scattered blocks between the pallets" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Scorch zones.", - "fixes": [] + "score": 4, + "reason": "Clean slab and blocks; no scorch or dust where the pallet was hit.", + "fixes": [ + "grime() scorch and dust around the debris pile and on the slab" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps the slab, inlay, tower discs and piles; debris reduces to a block.", + "fixes": [ + "keep two blocks per pile and the debris heap at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 5.25k / 1.95k / 476, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "73d1cc865dfce2f1be937b16a9a44ba9bfd47434aae6c0421550c06c2bef5f2a", - "sheet.png": "594fed08af9d356fce33ea5286b3245ac1570549a690e03c7d75a654a50d0269", - "raking.png": "a16a869d9bd8f4f676e35cc74d5b78ea92f18970764ef9d882bdad20252e54a1", - "raking_detail.png": "952128f8d18192614558c2b497096196a5d7a888fd7f135c80da6386fccba4ba", - "lods.png": "4df4fcdef9b34a3cd15cfc2099535aa3d1117cf8fc17ee050055b8bbc8eff7f7", - "silhouette.png": "853a8cb3a1f9e2fefdb654fe762a06e7f5eeec4415ed7b40a41547f49342050f", - "value.png": "a44496ab4f0d71def3a87beb7bd4200b382f8b4d3885bfd07ad851c208997ebe", - "detail.png": "429aaec98763567827565498e9d4d0865c95b4436309591c2b6fe973add57f76" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.989 - }, - { - "view": "side", - "iou": 0.88 - }, - { - "view": "3/4", - "iou": 0.952 - } - ], - "focal": { - "peakPart": "tower_r", - "peak": [ - 118, - 186 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "tower_r", - "share": 0.313 - }, - { - "part": "walls", - "share": 0.213 - }, - { - "part": "plinth", - "share": 0.166 - } - ], - "declared": [] - }, - "rest": { - "share": 0.732, - "spread": 0.721, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.123, - "straightShare": 0.771, - "taper": 0.741, - "shape": "stable" - } + "hero.png": "0499bdbff0f8edffc8d937121b102898359711a98455fdaa8740fa6eed7c450d", + "sheet.png": "e311a3eeeb022764f317e89597ab1880011c50b2f60c240b240b76149b88a826" } } diff --git a/recipes/rts_barracks/out/rts_barracks_foundation_rubble/critique.json b/recipes/rts_barracks/out/rts_barracks_foundation_rubble/critique.json index 998a5dae..7b447ea3 100644 --- a/recipes/rts_barracks/out/rts_barracks_foundation_rubble/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_foundation_rubble/critique.json @@ -1,105 +1,65 @@ { "asset": "rts_barracks_foundation_rubble", "reviewer": "agent", - "overall": 6.9, + "overall": 5.7, "scores": { "silhouette": { - "score": 6, - "reason": "Flattened site.", - "fixes": [] + "score": 5, + "reason": "Slab with tower discs, scattered cubes and fallen red slabs: reads as a wrecked site but very flat; nothing breaks the outline above 1 m.", + "fixes": [ + "one upright tipped pallet or a leaning timber stack so the ruin has a vertical" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "Same slab as the foundation; scattered blocks match pallet block sizes.", + "fixes": [ + "vary rubble block sizes (0.3-1.0x) so the scatter does not look like one repeated cube" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 5, + "reason": "Stone blocks and the slab read; the fallen team slabs are clean saturated red planes and the tower discs show a radial sliced-log pattern on top.", "fixes": [ - "window reveals / sills as their own stone" + "dusty, broken fallen slabs (fallenSlabs in site.ts)", + "tower footing tops as stone blocks, not a radial cap" ] }, "detail": { - "score": 6, - "reason": "Low stacked rubble, slabs, beams.", - "fixes": [] + "score": 5, + "reason": "Scattered blocks, two charred beams, fallen slabs; no small debris or splinters.", + "fixes": [ + "gravel and small shards around each block", + "splintered ends on the charred beams" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Rubble dust.", - "fixes": [] + "score": 4, + "reason": "Every broken block is a clean bevelled cube with crisp arrises; no soot, scorch or ash on the floor, no dust over the fallen team slabs, which stay bright and unmarked. Nothing says fire or collapse beyond the arrangement.", + "fixes": [ + "grime() scorch zones on wall stub tops and the floor around the heaps (rubble() in site.ts)", + "dust coat on the fallen team slabs (fallenSlabs in site.ts) so the red reads as buried, not freshly laid", + "chip/break one or two edges per rubble block instead of whole cubes" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 keeps the plinth, wall stubs and tower drums, but the tower drums turn into coarse octagons and the fallen team slabs merge into one flat red polygon, so the heap texture of LOD0 disappears at distance.", + "fixes": [ + "keep 10-12 sides on the tower stubs at LOD2 (a few dozen triangles)", + "keep 2-3 tilted heap blocks per bay at LOD2 instead of flattening the fallen slabs into one plane" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 4.88k / 1.97k / 388, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "452436bcd4a4084ca4cda0593e424f0072039280ff2142bab426842857bbf57a", - "sheet.png": "aedee4b2b3710e1b9d74ea9cb92e902b572306aa663c3e710e641cf22908f3d8", - "raking.png": "ee2d171ed1c33fdc82503024dc13a969ec466efe48479b351649a43984cb1155", - "raking_detail.png": "ecc58265b7e9347004913c2a5a4251db6df1ddef511572962b65902189344013", - "lods.png": "9e51f4190e4b830b532b6724b9f3db0ad244acb59e3d6414a2e77077d66cc808", - "silhouette.png": "1e20a2c9c871eff64fa7c0e8901fa64fedb2c8af70d12fc5ad83559947ca978e", - "value.png": "1086332be214360820b1a7fe3736838e5d2a59f948aeba236366de68478ae85e", - "detail.png": "9681d484ef01dee93953c7e9fa9071d51be55c03d035ca0d79922113dc03b44b" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.89 - }, - { - "view": "side", - "iou": 0.829 - }, - { - "view": "3/4", - "iou": 0.969 - } - ], - "focal": { - "peakPart": "walls", - "peak": [ - 350, - 250 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.456 - }, - { - "part": "plinth", - "share": 0.291 - }, - { - "part": "tower_r", - "share": 0.219 - } - ], - "declared": [] - }, - "rest": { - "share": 0.662, - "spread": 0.721, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.048, - "straightShare": 0.771, - "taper": 0.668, - "shape": "stable" - } + "hero.png": "b5c4590f3c2c616d5fcc2d18f79251ac66209bf82e5e7d58e3548fd35dbde787", + "sheet.png": "a4df8723f5e98f3706c6236385cca3d4c6fe4f8868f9e937ef37635c4f2e9319" } } diff --git a/recipes/rts_barracks/out/rts_barracks_rubble/critique.json b/recipes/rts_barracks/out/rts_barracks_rubble/critique.json index 86aafcff..5dcfedc7 100644 --- a/recipes/rts_barracks/out/rts_barracks_rubble/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_rubble/critique.json @@ -1,105 +1,65 @@ { "asset": "rts_barracks_rubble", "reviewer": "agent", - "overall": 7.0, + "overall": 5.7, "scores": { "silhouette": { - "score": 7, - "reason": "Wall stubs, stacked rubble heaps and fallen team slabs; ownership still reads (18.5% team share).", - "fixes": [] + "score": 5, + "reason": "A flat plinth with knee-high wall stubs, tower drum stubs and a heap of cubes over red roof slabs: reads as generic ruins; only the two drum stubs hint that it was this barracks. Front and side views are a low flat strip.", + "fixes": [ + "leave one gable corner or a tower stub standing 3-4 m high (plan.ts wallCells for rubble) so the ruin keeps a recognisable vertical" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "13.3 x 18.0 x 2.6 m on the same plinth and plan as the hall; block sizes match the standing walls.", + "fixes": [ + "vary rubble block sizes (0.3-1.0x) so the heap does not look like one repeated cube" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 5, + "reason": "Stone blocks read as stone, but the fallen roof slabs are clean saturated red sheets and the tower stub tops show a radial cut pattern that reads as a sliced log, not broken masonry.", "fixes": [ - "window reveals / sills as their own stone" + "break the fallen slabs into several pieces with darkened, dusty undersides (fallenSlabs in site.ts)", + "give tower stub tops a broken block surface (stepped, missing blocks) instead of the flat radial cap" ] }, "detail": { - "score": 6, - "reason": "Seeded block heaps stacked on face contacts, charred beams, fallen roof slabs, smoke.", - "fixes": [] + "score": 5, + "reason": "Wall stubs, block heaps, two charred beams and fallen slabs; blocks are near-identical bevelled cubes and there is no small debris (gravel, splinters, shards) between them.", + "fixes": [ + "scatter small debris and gravel around the heaps", + "more charred beams, some leaning on the wall stubs" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Rubble dust and chips on the heaps.", - "fixes": [] + "score": 4, + "reason": "Every broken block is a clean bevelled cube with crisp arrises; no soot, scorch or ash on the floor, no dust over the fallen team slabs, which stay bright and unmarked. Nothing says fire or collapse beyond the arrangement.", + "fixes": [ + "grime() scorch zones on wall stub tops and the floor around the heaps (rubble() in site.ts)", + "dust coat on the fallen team slabs (fallenSlabs in site.ts) so the red reads as buried, not freshly laid", + "chip/break one or two edges per rubble block instead of whole cubes" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 keeps the plinth, wall stubs and tower drums, but the tower drums turn into coarse octagons and the fallen team slabs merge into one flat red polygon, so the heap texture of LOD0 disappears at distance.", + "fixes": [ + "keep 10-12 sides on the tower stubs at LOD2 (a few dozen triangles)", + "keep 2-3 tilted heap blocks per bay at LOD2 instead of flattening the fallen slabs into one plane" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 10.3k / 3.66k / 592, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "61d261cc9576a9daf292e86a290939199b3f831a192025e17a0bdb1594365444", - "sheet.png": "327fce9f8103b9c5e72d8b043d52e44b531dcd5ad5c64742928c9ce29f27c71f", - "raking.png": "2cfec438a3611c8e67a7b4ed4b9fa08284037bb56065ec8e72344515975e10a6", - "raking_detail.png": "ae4a1e9be7cae5fcb05cffc48c565e7a30cd8b0ab217040bebc3f090665b4520", - "lods.png": "e3336c592ae3173db6797c3956fd237ad79c3c81c2d2083a0d9a53a10e3f3e7c", - "silhouette.png": "f2923d25c75bead08b098e1c34487fde0efef810acc7a76ca50673871ad78f03", - "value.png": "b6f1248e77d88a0b3e1873c8791616f65280c44dbfc8412e9c8033a1d3d324bd", - "detail.png": "bbd330561fc8b2cdf31697a9121aa0d6050c7498b34e82daa9fc18610bb69105" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.878 - }, - { - "view": "side", - "iou": 0.843 - }, - { - "view": "3/4", - "iou": 0.953 - } - ], - "focal": { - "peakPart": "walls", - "peak": [ - 209, - 216 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.453 - }, - { - "part": "rubble", - "share": 0.197 - }, - { - "part": "plinth", - "share": 0.164 - } - ], - "declared": [] - }, - "rest": { - "share": 0.472, - "spread": 0.739, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.026, - "straightShare": 0.729, - "taper": 0.627, - "shape": "stable" - } + "hero.png": "2dc3a96e1c0effffe10e1ee87c65c80a372ef08ef246e663588a1f29a49f9973", + "sheet.png": "16f96b33047d909ec0d00bf97240b27c71d729ffadff5ddb056fcba8f930d1c0" } } diff --git a/recipes/rts_barracks/out/rts_barracks_scaffold/critique.json b/recipes/rts_barracks/out/rts_barracks_scaffold/critique.json index 08e651ef..eb139d35 100644 --- a/recipes/rts_barracks/out/rts_barracks_scaffold/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_scaffold/critique.json @@ -1,105 +1,65 @@ { "asset": "rts_barracks_scaffold", "reviewer": "agent", - "overall": 7.1, + "overall": 6.4, "scores": { "silhouette": { "score": 7, - "reason": "Half-built walls inside a scaffold; team tarp and banner carry the read.", - "fixes": [] + "reason": "Half-height walls stepped at the top, tower drums without caps, a scaffold of poles, ledgers and pine plank decks, and two banners: reads clearly as a barracks under construction.", + "fixes": [ + "a hoist or gin-pole with a rope on one tower so the build stage reads from far away" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "15.6 x 18.0 x 5.5 m; scaffold lifts at a believable ~3.2 m deck height; poles look slightly too thin for timber scaffold at this scale.", + "fixes": [ + "scaffold poles ~0.12 m diameter (site.ts scaffold()) instead of the current thin tubes" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 6, + "reason": "Pine decks read as fresh wood, stone blocks as stone; poles read as grey metal tubes rather than weathered timber poles, and in sheet.png the plank decks render pale grey-white.", "fixes": [ - "window reveals / sills as their own stone" + "darker, grained wood_weathered poles with lashings at the nodes instead of the black clamp blocks", + "check the pine role on deck planks: pale grey in sheet.png while wood in hero.png" ] }, "detail": { - "score": 7, - "reason": "Poles, ledgers with steel couplers at every crossing, pine plank decks, team tarp, build dust.", - "fixes": [] + "score": 6, + "reason": "Poles, ledgers, braces, plank decks, stepped wall tops and an inlay inside; no ladders, ropes, buckets or loose blocks on the decks.", + "fixes": [ + "a ladder between deck and ground", + "blocks and a mortar bucket on the decks", + "rope lashings at pole nodes" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy.", - "fixes": [] + "score": 5, + "reason": "Clean stone and fresh decks; no mortar drips down the walls, no mud at the pole feet.", + "fixes": [ + "mortar drips below the top courses and splash on the plinth", + "mud at pole feet" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 keeps walls, towers, poles and banners, but the plank decks disappear and the walls lose their coursing, so the scaffold reads as bare sticks.", + "fixes": [ + "keep the plank decks as single quads at LOD2 (site.ts scaffold())", + "value stripes for coursing at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 17.9k / 4.59k / 704, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "89cea2458f17fbfec6cd4d855b5b325b9a0255159b1a5710244bb32f5f445248", - "sheet.png": "b5c6a81e566a6fac97d7b96c4a83c7439a77a04b916f32e79540376f3638e852", - "raking.png": "077d9d650b706ae7603f31be6ef6318a7ddeb52a2f99294a8c700f8e443218a3", - "raking_detail.png": "4f27219d2033bda16726654fa48adb715eb006d53f0e8de1a5e70ae1fea19599", - "lods.png": "8c2c63ff1c621ca765a4e43afe5b90aa1bd78f142d94c016fbc179eb5d6552ad", - "silhouette.png": "e54fa297180170ad62c2b38496c1f3d40b955c694c9b0ce51f9615099141000a", - "value.png": "0d59bbedd4b4aa887fb398063bc54213c1c0d26ed8cc34ee2af87646a5a409b2", - "detail.png": "1340de824e10e62844e350b38bdc008fcfa44f2d31eb98f4fb33ff60144644a8" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.761 - }, - { - "view": "side", - "iou": 0.848 - }, - { - "view": "3/4", - "iou": 0.974 - } - ], - "focal": { - "peakPart": "tower_r", - "peak": [ - 179, - 133 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "scaffold", - "share": 0.351 - }, - { - "part": "walls", - "share": 0.157 - }, - { - "part": "banner_r", - "share": 0.151 - } - ], - "declared": [] - }, - "rest": { - "share": 0.396, - "spread": 0.787, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.441, - "straightShare": 0.781, - "taper": 0.693, - "shape": "stable" - } + "hero.png": "f71898beceb9aea5a64f2e12e31d1bb98b733b6243eac8a196155662a91f0872", + "sheet.png": "60fa7bf0b44041f603e8911649b578047d3f2ca3549e5f17838d4a2b75bd6e8e" } } diff --git a/recipes/rts_barracks/out/rts_barracks_scaffold_damaged/critique.json b/recipes/rts_barracks/out/rts_barracks_scaffold_damaged/critique.json index 085394b7..47d7b6bf 100644 --- a/recipes/rts_barracks/out/rts_barracks_scaffold_damaged/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_scaffold_damaged/critique.json @@ -1,105 +1,60 @@ { "asset": "rts_barracks_scaffold_damaged", "reviewer": "agent", - "overall": 7.1, + "overall": 6.4, "scores": { "silhouette": { "score": 7, - "reason": "Scaffolded walls with knocked holes.", - "fixes": [] + "reason": "Scaffold stage with notched walls, a breach in the front wall and a debris heap at the front: reads as damaged construction; the scaffold itself is intact.", + "fixes": [ + "one collapsed scaffold bay (fallen ledger and dangling plank) so the damage reads in outline" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "Same 15.6 x 18.0 x 5.5 m as the scaffold stage; debris sizes match the wall blocks.", + "fixes": [ + "scaffold poles ~0.12 m diameter instead of thin tubes" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 6, + "reason": "As the scaffold stage: pine decks read as wood, poles as grey metal, stone as stone.", "fixes": [ - "window reveals / sills as their own stone" + "wood_weathered poles with lashings instead of black clamp blocks" ] }, "detail": { - "score": 7, - "reason": "As scaffold plus holes, debris heap, scorch, smoke and fire.", - "fixes": [] + "score": 6, + "reason": "Scaffold detail plus wall notches and a debris heap; no broken planks or bent poles.", + "fixes": [ + "a split plank and a leaning pole near the breach" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Scorch zones and rubble dust.", - "fixes": [] + "score": 5, + "reason": "Debris tells the story; no scorch, dust or soot around the breach.", + "fixes": [ + "grime() scorch and dust spray around the breach and the debris heap" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 drops the plank decks and coursing, as on the scaffold stage.", + "fixes": [ + "keep the plank decks as single quads at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 16.9k / 4.79k / 728, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "7b5f3460e05ddf59c824679d04fff5a7744684d1c8a51a4d763b28b93ad7702c", - "sheet.png": "0d78a88c8a0e1df63c194add50bdb3bb203ed81166e16ef172744099d8c0dd9a", - "raking.png": "61cd9a4f1095119cfb28fb6a4b8afd9783338f971a2ace64c1ea7dacdc75f904", - "raking_detail.png": "bb5c1d992f38339ede44185843d7abe54a3a5c9e426cc9ee1c73a82848bc9f96", - "lods.png": "d82ff1ef72682a4e53dcebe3da411955e7312b21c04f1e1e8a111e1b3a668928", - "silhouette.png": "bd06a79ff3e76d03d02e94f50dd30ab555627b1129bc6624b0557978e379bafe", - "value.png": "d5840e42a9c6ff9a8bcb085c2895650c40cd665e234afcd8501d0e84d1717fed", - "detail.png": "315951652c25fa74c922006e6313c88624c7594e440612e67a131b3f5cd5d871" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.798 - }, - { - "view": "side", - "iou": 0.895 - }, - { - "view": "3/4", - "iou": 0.965 - } - ], - "focal": { - "peakPart": "tower_r", - "peak": [ - 175, - 133 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "scaffold", - "share": 0.273 - }, - { - "part": "banner_r", - "share": 0.146 - }, - { - "part": "tower_r", - "share": 0.114 - } - ], - "declared": [] - }, - "rest": { - "share": 0.406, - "spread": 0.786, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.494, - "straightShare": 0.775, - "taper": 0.701, - "shape": "aggressive" - } + "hero.png": "68cd2dd9da789c6eaecfa397f927d902401005e590e5f4d14b0768eda0b46f5f", + "sheet.png": "79294122b16a039bc8c4884809fbf317c2012796ed90ad04752a08442d98455c" } } diff --git a/recipes/rts_barracks/out/rts_barracks_scaffold_rubble/critique.json b/recipes/rts_barracks/out/rts_barracks_scaffold_rubble/critique.json index 6df47563..fea8b762 100644 --- a/recipes/rts_barracks/out/rts_barracks_scaffold_rubble/critique.json +++ b/recipes/rts_barracks/out/rts_barracks_scaffold_rubble/critique.json @@ -1,105 +1,65 @@ { "asset": "rts_barracks_scaffold_rubble", "reviewer": "agent", - "overall": 6.9, + "overall": 5.7, "scores": { "silhouette": { - "score": 6, - "reason": "Low stubs and heaps; reads as a ruin.", - "fixes": [] + "score": 5, + "reason": "Low wall courses around a heap of cubes over red slabs, two tower stubs and the gate steps: reads as ruins, flat in front and side views; the scaffold is gone entirely.", + "fixes": [ + "leave a few fallen scaffold poles and a leaning ledger in the ruin so it still reads as the scaffold stage" + ] }, "proportions": { "score": 7, - "reason": "10 x 14 m hall, 5.1 m eaves, 3.5 m gate (measures gate) = 1.8 x a Warden; storey and door scale per the house playbook.", - "fixes": [] + "reason": "Same plinth and plan as the scaffold stage; rubble blocks match the wall blocks.", + "fixes": [ + "vary rubble block sizes (0.3-1.0x)" + ] }, "materials": { - "score": 7, - "reason": "Dressed-stone blocks (concrete grain at LOD0, the stone strip's coursing at LOD1/2), coursed drum towers (chamfered bed joints in the profile), lime-plaster gables in an oak frame, oak porch posts and rafters, tarred oak gate, team slate roof lapped like real sheets, iron fittings, brass.", + "score": 5, + "reason": "Stone and plinth read; fallen team slabs are clean red planes and tower tops show the radial sliced pattern.", "fixes": [ - "window reveals / sills as their own stone" + "dusty, broken fallen slabs (fallenSlabs in site.ts)", + "tower stub tops as stepped broken blocks" ] }, "detail": { - "score": 6, - "reason": "Stacked rubble, charred beams, fallen slabs.", - "fixes": [] + "score": 5, + "reason": "Wall stubs, block heaps and fallen slabs; no planks, poles or small debris from the scaffold.", + "fixes": [ + "charred planks and poles from the scaffold in the heap", + "gravel and shards between blocks" + ] }, "wear": { - "score": 6, - "reason": "weather(): chipped arrises 10%, moss + damp on 35% of the 1.1 m wall foot, salt dust on ledges, grime streaks running 1.6 m down from eaves drips and window sills (gravity-traced); timber dusty/muddy. Rubble dust.", - "fixes": [] + "score": 4, + "reason": "Every broken block is a clean bevelled cube with crisp arrises; no soot, scorch or ash on the floor, no dust over the fallen team slabs, which stay bright and unmarked. Nothing says fire or collapse beyond the arrangement.", + "fixes": [ + "grime() scorch zones on wall stub tops and the floor around the heaps (rubble() in site.ts)", + "dust coat on the fallen team slabs (fallenSlabs in site.ts) so the red reads as buried, not freshly laid", + "chip/break one or two edges per rubble block instead of whole cubes" + ] }, "lod": { - "score": 8, - "reason": "LOD2 one box per bay from the same cell plan, so stages and damage agree at every LOD; <= 900 tris.", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 keeps the plinth, wall stubs and tower drums, but the tower drums turn into coarse octagons and the fallen team slabs merge into one flat red polygon, so the heap texture of LOD0 disappears at distance.", + "fixes": [ + "keep 10-12 sides on the tower stubs at LOD2 (a few dozen triangles)", + "keep 2-3 tilted heap blocks per bay at LOD2 instead of flattening the fallen slabs into one plane" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Banners hang from a sewn sleeve on a bracketed crossbar; rubble stacks block-on-block.", - "fixes": [] + "reason": "QA PASS on every gate (rebuilt today), tris 9.88k / 3.29k / 580, draws 2 / 2 / 2; deterministic, one trim material + team slot. Triangles mostly sit in the running-bond blocks, where they show.", + "fixes": [ + "merge the per-block wall boxes into one course mesh per bay at LOD0 (shared inner faces removed) to cut the block-count triangles without losing the coursing" + ] } }, "renders": { - "hero.png": "3e0d52b680c2eed2f0b197ac6ba75a3bd4dc4608eaee02ed925ce8a22ead8506", - "sheet.png": "74411f0b47ec12e970e7779a20d1c103b766f432867c5f3ef1832756882b345e", - "raking.png": "581a7792d158e87f1a929d694151e36c01aec3e832995f324b0763de7220d2e8", - "raking_detail.png": "d4ff5ebde4f2bccdffd0f0f5a590648f6dfcd175fcc0451f62d65a429716a91b", - "lods.png": "44a3ed1567aeed4d35b80791c110e8a760e88070b6ca3aabbee75df1e4382c30", - "silhouette.png": "1b73369eb7d0e4bd88a293aba168acbbe8fc84403d10166b8d536ae1982bff9b", - "value.png": "4ffeced28694ae339cccf06368a03d5bc1cd60c32bfd1a77119dfc24576367bb", - "detail.png": "af39044803c730081420a79610cc6dcf53f7b3d75b6ac6a39a87cf5719c350df" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.907 - }, - { - "view": "side", - "iou": 0.956 - }, - { - "view": "3/4", - "iou": 0.969 - } - ], - "focal": { - "peakPart": "walls", - "peak": [ - 213, - 204 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walls", - "share": 0.553 - }, - { - "part": "plinth", - "share": 0.199 - }, - { - "part": "rubble", - "share": 0.099 - } - ], - "declared": [] - }, - "rest": { - "share": 0.457, - "spread": 0.76, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.048, - "straightShare": 0.79, - "taper": 0.489, - "shape": "stable" - } + "hero.png": "46a9674f9a1e8dd268ff8ea7acae50cfcfc21dfdacdbd0585ad96a122154ac4a", + "sheet.png": "d7a45d8a827f808ef64374e75d0b6b190c5bb254a722bae5e8aba2a7b07b3d2e" } } diff --git a/recipes/rts_faction/out/rts_faction/critique.json b/recipes/rts_faction/out/rts_faction/critique.json index c61d9c1e..f2f5a9bf 100644 --- a/recipes/rts_faction/out/rts_faction/critique.json +++ b/recipes/rts_faction/out/rts_faction/critique.json @@ -1,105 +1,61 @@ { "asset": "rts_faction", "reviewer": "agent", - "overall": 7.6, + "overall": 6.6, "scores": { "silhouette": { - "score": 8, - "reason": "The roster reads as one faction: bone/team/amber palette and chevron language across hall, units, tank and deposit.", - "fixes": [] + "score": 7, + "reason": "Barracks, a squad of soldiers, a tank and a crystal deposit on one ground plate: the roster reads as one faction (bone/team red/amber), and every unit is identifiable in the 3/4 views. As a composition it is a line-up, not a base layout.", + "fixes": [ + "arrange the units with intent (squad in formation at the gate, tank parked on its path) and add a perimeter element such as a low wall or fence segment" + ] }, "proportions": { "score": 8, - "reason": "All assets at true scale via ctx.use: Wardens fit the 3.5 m gate, the tank is ~3 units long.", - "fixes": [] + "reason": "30 x 25 x 11.2 m plate; all assets come in at true scale through ctx.use, soldiers fit the 3.5 m gate, tank ~3 soldiers long.", + "fixes": [ + "scale check against a real compound: a 30 x 25 m plate is tight for a barracks plus a tank park" + ] }, "materials": { - "score": 7, - "reason": "Inherits every asset's strips; salt-flat sand ground with gravel paths to the gate and the tank park.", - "fixes": [] + "score": 6, + "reason": "Each asset keeps its strips; the ground plate meant as salt-flat sand reads as light wood planks (straight grain stripes) in hero.png, and the two paths are barely distinguishable from it.", + "fixes": [ + "ground plate on a sand/soil strip (not one with straight grain) with gravel paths that contrast in value and roughness (rts_faction.ts)" + ] }, "detail": { - "score": 7, - "reason": "Lineup on a ground plate with two trodden paths.", + "score": 5, + "reason": "Hall, squad, tank and deposit on an empty plate with two paths; nothing ties them together (no crates, fuel drums, fences, signage, tyre tracks).", "fixes": [ - "props between assets (crates, a fuel drum)" + "props between assets: crates, fuel drums, a sandbag line", + "tyre tracks from the tank park to the gate" ] }, "wear": { - "score": 7, - "reason": "Inherits every asset's measured wear (prefixed through ctx.use).", - "fixes": [] + "score": 6, + "reason": "Inherits each asset's weather(); the plate itself is clean, with no tracks, puddles or trampled ground around the gate.", + "fixes": [ + "grime() trampled darker ground at the gate and track ruts behind the tank" + ] }, "lod": { - "score": 8, - "reason": "LOD2 3.6k tris for the whole roster.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 (3.66k tris) keeps every asset, colour block and the team red; the tank loses its wheels to a belt slab and soldiers become blocky.", + "fixes": [ + "keep the tank wheel discs at LOD2 (rts_tank running.ts) so the parked tank does not change outline" + ] }, "technical": { - "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18.", - "fixes": [] + "score": 7, + "reason": "QA PASS (rebuilt today): 143k / 31.2k / 3.66k tris, 36 / 36 / 22 draws. Fine for a showcase, but 36 draws at LOD0 is heavy for a single placed asset.", + "fixes": [ + "merge static parts across the composed assets into one body per material so LOD0 drops well below 36 draws" + ] } }, "renders": { - "hero.png": "4831ea284891fccfe6bd8efbbc3903b545b4de061d58bf3b4f68eab7ad8316aa", - "sheet.png": "560bcfc9210718367154b969c03681f19adfa62675266b1f2a6c8d71ecb53197", - "raking.png": "0dde5215d50effbf9fd771e2611fe370ed67c47abce4477b0628792c7002d0ec", - "raking_detail.png": "7ba7733a242fa1eac6abfcfb866cc65ddadd279b05ad3be31df1e2a31fc76995", - "lods.png": "9b3136137ef5ff78fff86c37e4e7711685d0e17b63e55db1bd89d1cbeac56d64", - "silhouette.png": "188f6f0f189d5178cb1a2c5723d3d9d4f2a1a2848651906f8da27e73b00c9377", - "value.png": "1354949fb6ca2f3dc00e5d883f3c879d699073e2823df085c2738c8fff7f5805", - "detail.png": "6473c03ccd80a21fbbd7baa0555f3840e7d5a0f6285af82c6174b6b4c5b9fe4d" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.985 - }, - { - "view": "side", - "iou": 0.969 - }, - { - "view": "3/4", - "iou": 0.991 - } - ], - "focal": { - "peakPart": "tank_hull", - "peak": [ - 280, - 299 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "ground", - "share": 0.376 - }, - { - "part": "tank_hull", - "share": 0.127 - }, - { - "part": "hall_plinth", - "share": 0.101 - } - ], - "declared": [] - }, - "rest": { - "share": 0.556, - "spread": 0.861, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.458, - "straightShare": 0.58, - "taper": 0.777, - "shape": "aggressive" - } + "hero.png": "920d81c530c7754a0c679d2110f5cec47b9c987fb14b2b69286d90cdfcec4db7", + "sheet.png": "1d74e0da8c9e5e019af5c05e6126d0dfb3d8c632df0243c8b80573433224bea7" } } diff --git a/recipes/rts_resource/out/rts_resource/critique.json b/recipes/rts_resource/out/rts_resource/critique.json index 65e195c2..0dec99cd 100644 --- a/recipes/rts_resource/out/rts_resource/critique.json +++ b/recipes/rts_resource/out/rts_resource/critique.json @@ -1,107 +1,63 @@ { "asset": "rts_resource", "reviewer": "agent", - "overall": 7.1, + "overall": 6.3, "scores": { "silhouette": { - "score": 8, - "reason": "Dark outcrop with tall amber crystal clusters and a team claim stake: a resource at any distance.", - "fixes": [] + "score": 7, + "reason": "Dark rounded outcrop with amber crystal clusters spiking out and a red claim flag: reads as a resource node at any distance. The rock itself is a smooth blob; the crystals stick out almost horizontally and look pinned on.", + "fixes": [ + "grow the crystal clusters out of fracture pockets in the rock (seat them 20-30% deep) and angle them upward" + ] }, "proportions": { "score": 7, - "reason": "4.9 x 4.4 x 3.6 m: one crew's mining site beside 1.8 m workers.", - "fixes": [] + "reason": "4.9 x 4.4 x 3.6 m: a crew-sized mining site beside 1.8 m workers. Crystals up to ~1.5 m are big but fit an RTS node.", + "fixes": [ + "shrink the side clusters ~20% so the top cluster leads the composition" + ] }, "materials": { - "score": 6, - "reason": "SDF basalt on the concrete strip (jointed strips show contour lines on marching-cubes UVs), bedding bands on steep faces, salt crust on up-faces, crystals ramped deep amber -> pale gold.", + "score": 5, + "reason": "Basalt reads dark, but the salt crust is blotchy white patches like camouflage, horizontal contour banding from the SDF UVs shows in the side views, and the crystals are flat amber plastic with no translucency, inner glow or facet sheen.", "fixes": [ - "texture-space rock strip that survives SDF UVs" + "texture-space rock strip that survives SDF UVs (triplanar-like box projection) to remove the contour banding", + "crystal material with emissive core and lighter facet edges instead of flat amber" ] }, "detail": { - "score": 7, - "reason": "Three fracture planes, four crystal clusters, claim stake with finial, sleeved pennant and plaque.", - "fixes": [] + "score": 5, + "reason": "Four crystal clusters, a claim stake with finial, pennant and a flat red ground plate; the rock has no visible fracture planes, ledges or strata at hero distance.", + "fixes": [ + "sharp fracture facets and a couple of ledges on the outcrop (SDF subtract planes)", + "small loose crystal shards at the base" + ] }, "wear": { - "score": 6, - "reason": "weather(): salt dust on 30% of up-faces, brine mud on 50% of the ground band; no mining scars on the full node.", + "score": 5, + "reason": "Salt dust on top faces and brine mud at the base exist but read as random patches; no mining scars, pick marks or spoil.", "fixes": [ - "pick marks near the mine points" + "pick marks and a small spoil heap next to one cluster", + "salt dust concentrated in crevices rather than blotches" ] }, "lod": { - "score": 8, - "reason": "LOD2 486 tris keeps the outcrop and crystals; team share 12.9%.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps the outcrop, all crystal clusters and the flag; salt patches become big flat facets.", + "fixes": [ + "lower-contrast salt at LOD2 so the facets do not flash" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Pennant flies from a hem sleeve round the pole (face contact).", - "fixes": [] + "reason": "QA PASS (rebuilt today): 12.3k / 2.96k / 486 tris, 3 / 3 / 3 draws; deterministic, one trim material + team slot. Most LOD0 triangles go to the smooth SDF rock.", + "fixes": [ + "coarser SDF cell on the rock body, finer on the crystal bases" + ] } }, "renders": { - "hero.png": "c73fc3d7daddbe08a5e0db8bf6ad3add6733f83ec4ded88f9e572ac03654e377", - "sheet.png": "5bd03450c0876133b7efba0253a40ce2a241a15d357abd5fa414ae639e807a88", - "raking.png": "d94442b7c89d1aa28d98a999ab09fa17f622fcd1f417957aa9211ea80503ad1c", - "raking_detail.png": "1381aed13053a7cccf608b74701d7e78db860eb8e29ac34e615770cbb3ea3725", - "lods.png": "2f230e12d9c49c9acc9e05f0f1c110487b11df058c1dc4cddcf15237431b4afb", - "silhouette.png": "8c96d2d87bed6e9b6ff7b81bff05e2342aa0be9ccd330a44604e57477bb140ef", - "value.png": "2cfa4b37b51365e9b5dc9cbf9b85f2f03084738eeec358aadb695d2ead949240", - "detail.png": "e569b46614ba831db3d36f6346c9cfe329ec3147ab38577f806baf29b857cb33" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.807 - }, - { - "view": "side", - "iou": 0.862 - }, - { - "view": "3/4", - "iou": 0.796 - } - ], - "focal": { - "peakPart": "rock", - "peak": [ - 362, - 164 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "crystals", - "share": 0.477 - }, - { - "part": "rock", - "share": 0.305 - }, - { - "part": "claim_pennant", - "share": 0.173 - } - ], - "declared": [] - }, - "rest": { - "share": 0.386, - "spread": 0.557, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.581, - "straightShare": 0.476, - "taper": 0.742, - "shape": "aggressive" - } + "hero.png": "4c91792a85bcbe4d4d72e2a0ba18b72e4362c905a9e17b931f8c89d56713fa87", + "sheet.png": "83468a77e0300a01feeb999d1a9d9411ea73966c00d811b2fee0768de178e32e" } } diff --git a/recipes/rts_resource/out/rts_resource_depleted/critique.json b/recipes/rts_resource/out/rts_resource_depleted/critique.json index 06ebecad..9913c125 100644 --- a/recipes/rts_resource/out/rts_resource_depleted/critique.json +++ b/recipes/rts_resource/out/rts_resource_depleted/critique.json @@ -1,105 +1,62 @@ { "asset": "rts_resource_depleted", "reviewer": "agent", - "overall": 7.0, + "overall": 6.1, "scores": { "silhouette": { - "score": 7, - "reason": "Scooped quarry and sheared stubs: reads harvested; team stake keeps ownership (14.5%).", - "fixes": [] + "score": 6, + "reason": "The same outcrop with a scooped quarry top and sheared crystal stubs, plus the claim flag: reads as harvested, but without crystals it is an anonymous rounded boulder from the side.", + "fixes": [ + "keep one or two broken crystal shards standing (sheared halfway, jagged) so the node stays identifiable" + ] }, "proportions": { "score": 7, - "reason": "4.9 x 4.4 x 3.6 m: one crew's mining site beside 1.8 m workers.", - "fixes": [] + "reason": "4.9 x 3.8 x 2.7 m, lower than the full node: plausible after harvesting.", + "fixes": [ + "make the quarry scoop deeper (~0.8 m) relative to the rim" + ] }, "materials": { - "score": 6, - "reason": "SDF basalt on the concrete strip (jointed strips show contour lines on marching-cubes UVs), bedding bands on steep faces, salt crust on up-faces, crystals ramped deep amber -> pale gold.", + "score": 5, + "reason": "Basalt with blotchy salt patches and SDF contour banding as on the full node; the sheared crystal stubs read as brown wood stumps, not broken amber crystal.", "fixes": [ - "texture-space rock strip that survives SDF UVs" + "crystal stubs keep the amber crystal colour with a pale fractured top", + "texture-space rock strip without contour banding" ] }, "detail": { - "score": 7, - "reason": "Quarry scoop, sheared crystal stubs, spoil dust emitter.", - "fixes": [] + "score": 5, + "reason": "Quarry scoop, crystal stubs and the flag; no tool marks, spoil or rubble in the scoop.", + "fixes": [ + "spoil and crystal chips inside the scoop", + "pick and drill marks on the scoop walls" + ] }, "wear": { - "score": 6, - "reason": "Same measured salt dust and mud as the full node; the scoop is the mining scar.", - "fixes": [] + "score": 5, + "reason": "The scoop is the scar; salt and mud patches as on the full node. No dust or spoil around the base.", + "fixes": [ + "spoil dust fan around the base on the side the crew worked from" + ] }, "lod": { - "score": 8, - "reason": "LOD2 486 tris keeps the outcrop and crystals; team share 12.9%.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 keeps the outcrop, scoop, stubs and flag; salt becomes big flat facets.", + "fixes": [ + "lower-contrast salt at LOD2" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Pennant flies from a hem sleeve round the pole (face contact).", - "fixes": [] + "reason": "QA PASS (rebuilt today): 12.6k / 2.98k / 474 tris, 2 / 2 / 2 draws.", + "fixes": [ + "coarser SDF cell on the rock body" + ] } }, "renders": { - "hero.png": "80dabd0af2c1184b3c5abf9e38e319ffcdaf767ada36177d7f74f9aa3ff316eb", - "sheet.png": "ea1fbfd329b87b16e35c8da378819ae8ccec728e57bcd1b80d08a4362eedd625", - "raking.png": "ee58ea4dd73fc65dc0900b96777d9d0b9c9a1b49c674a6583590a904e23089ad", - "raking_detail.png": "b2254bd9a9f5cda924acf183889d4524aafcd22c44a65dbd9ab31953056b3c3b", - "lods.png": "2dbbdbd0a566c794f04cf641b6659accc79a3a16abddd594fbe3239b2da2d2d3", - "silhouette.png": "5bcfd526b009880348386018f3e976ed3ed4a84e3b0267293d9dbf2614978b42", - "value.png": "000fdc25186e148ab809658cbe4996d9d2340c76413afa79d3c2498cf39e4ee2", - "detail.png": "eab5ba17829e910eef43498e8304f6dcaed1ca6177d5f6eb13c52be9af8f3631" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.844 - }, - { - "view": "side", - "iou": 0.857 - }, - { - "view": "3/4", - "iou": 0.832 - } - ], - "focal": { - "peakPart": "rock", - "peak": [ - 370, - 200 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "rock", - "share": 0.871 - }, - { - "part": "claim_pole", - "share": 0.102 - }, - { - "part": "claim_pennant", - "share": 0.027 - } - ], - "declared": [] - }, - "rest": { - "share": 0.514, - "spread": 0.628, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.464, - "straightShare": 0.602, - "taper": 0.499, - "shape": "aggressive" - } + "hero.png": "4327661f4ec4f547da7975d0c5504fa49c283b8a9d66bb967761f48b7442affa", + "sheet.png": "0d1f367378855902cbaa32a6740e9e53c461b7d08e69a680bf0a2b946031f2d2" } } diff --git a/recipes/rts_soldier/out/rts_soldier/critique.json b/recipes/rts_soldier/out/rts_soldier/critique.json index 6316cb23..c2b05e7e 100644 --- a/recipes/rts_soldier/out/rts_soldier/critique.json +++ b/recipes/rts_soldier/out/rts_soldier/critique.json @@ -1,127 +1,64 @@ { "asset": "rts_soldier", "reviewer": "agent", - "overall": 7.0, + "overall": 6.4, "scores": { "silhouette": { "score": 7, - "reason": "Chevron crest, big team pauldrons and the carbine read 'infantry' at 64 px; IoU 0.94/0.91/0.90.", + "reason": "Red crest, big red pauldrons, carbine and backpack radio read \"infantry\" in the 3/4 views; in the front view the carbine points at the camera and nearly vanishes, and the stiff straight-arm stance is generic.", "fixes": [ - "longer crest tail for the back view" + "carry the carbine at port arms across the chest so it reads from the front as well", + "a longer crest tail for the back view" ] }, "proportions": { "score": 7, - "reason": "Same 6-head figure as the Delver; carbine sized to the fist (grip joint).", - "fixes": [] + "reason": "1.3 x 0.8 x 2.0 m, a chunky ~6-head figure; legs and pauldrons read at RTS scale. The head and hands are small next to the pauldrons for a 64 px read.", + "fixes": [ + "head and hands ~10% bigger for the RTS read" + ] }, "materials": { - "score": 7, - "reason": "Ceramic plastic armour and team plates, canvas undersuit, leather gloves/boots/belt, painted gunmetal carbine with leather grip, aged brass collar and muzzle.", + "score": 6, + "reason": "Bone ceramic armour, olive canvas undersuit, brown leather boots and gunmetal carbine separate in hero.png; in sheet.png the boots and gloves render pale blue-grey/white, and every surface has the same soft plastic sheen.", "fixes": [ - "armour panel lines on the torso shell" + "rougher matte canvas versus glossier ceramic armour (roughness contrast)", + "check the leather role on boots/gloves: pale blue-grey in sheet.png" ] }, "detail": { - "score": 6, - "reason": "Visor housing, brow ridge, ear guards, chest plates, abdomen bands, belt pouches, knee/shin guards, cooling rings, power pack vents and whip antenna.", + "score": 5, + "reason": "Visor, crest, chest plates, red stripe, knee pads, belt and radio pack with antenna; the body is blocky low-poly limbs with no webbing, pouches or panel lines visible at hero distance.", "fixes": [ - "webbing straps over the chest", - "magazine pouches" + "webbing straps and magazine pouches on the chest", + "panel lines on the torso shell", + "rounder limb cross-sections (8 sides) at LOD0" ] }, "wear": { - "score": 6, - "reason": "weather(): chips on 14% of hard convex edges, salt dust on 30% of up-faces, brine mud on 50% of the ground band (boots, shins); cloth/straps dust+mud only. Measured and gated. Subtle at hero distance; LOD2 colour still carries the read.", + "score": 5, + "reason": "weather() chips, dust and mud are measured, but at hero distance the figure reads clean; boots carry no visible mud.", "fixes": [ - "scuffs on knee guards visible at hero distance" + "stronger mud on boots and shins, scuffs on knee pads visible at hero distance" ] }, "lod": { - "score": 8, - "reason": "LOD2 380 tris at the crowd budget; team share 27.1%.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 (348 tris) keeps crest, pauldrons, carbine and team colour; boots become blue-grey blocks.", + "fixes": [ + "keep the boot colour consistent between LODs" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18.", - "fixes": [] + "reason": "QA PASS (rebuilt today): 12.1k / 2.6k / 348 tris, 3 / 3 / 3 draws; triangles mostly on the head and carbine.", + "fixes": [ + "spend fewer LOD0 triangles on the carbine barrel rings and more on rounder limbs" + ] } }, "renders": { - "hero.png": "d917e12d9fec1dafe5675c108bc44fa63a8412dd8622eda09a3209ee564329fc", - "sheet.png": "16c5e46e4ace7877c15464e0a5a830156de2ffa19588973100382aafde07c5b7", - "raking.png": "c1a8921df65b5682b9155aaf08dfc1bc5a67dbb31d8d319bce7136127f0e35e0", - "raking_detail.png": "18c4333326bf833e7427246d342e685e95f308d94dcfdd8c1ae37dffd6696b09", - "lods.png": "d84c9b63984375c873a77b0bc50d0b6f7d7660513a78877a8b9a30a496713c9c", - "silhouette.png": "21d6f26e67f1b3d209c730a4a880456df3189640794d32b8507bd5453d87b026", - "value.png": "3ac9b7f9c1f053e669eb22e2dcf9a4a523486d468c80bfcbe605236fc8a0aa2b", - "detail.png": "6bed7e1cbb87a861ef53fda14f6c3ac23c5fddb542610b662a667adce990d9bb" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.939 - }, - { - "view": "side", - "iou": 0.906 - }, - { - "view": "3/4", - "iou": 0.903 - } - ], - "focal": { - "peakPart": "helmet", - "peak": [ - 378, - 77 - ], - "focalShare": 0.406, - "focalArea": 0.089, - "top": [ - { - "part": "helmet", - "share": 0.214 - }, - { - "part": "torso", - "share": 0.181 - }, - { - "part": "crest", - "share": 0.116 - } - ], - "declared": [ - "helmet", - "visor", - "crest" - ] - }, - "rest": { - "share": 0.211, - "spread": 0.564, - "min": 0.2 - }, - "grounded": { - "rise": 0.244, - "bands": [ - 0.225, - 0.282, - 0.254, - 0.37, - 0.425 - ] - }, - "shape": { - "acuteShare": 0.425, - "straightShare": 0.722, - "taper": 0.326, - "shape": "stable", - "declared": "stable" - } + "hero.png": "5d80d58ae7fe2b60c3937cec7d4c643583187247e14f86529bb73c43864155ed", + "sheet.png": "0fb2a5405529051675f7f9368b171fce273e95e28a6f04b6abd26236555aae72" } } diff --git a/recipes/rts_tank/out/rts_tank/critique.json b/recipes/rts_tank/out/rts_tank/critique.json index d83481f4..1855e19c 100644 --- a/recipes/rts_tank/out/rts_tank/critique.json +++ b/recipes/rts_tank/out/rts_tank/critique.json @@ -1,116 +1,63 @@ { "asset": "rts_tank", "reviewer": "agent", - "overall": 7.3, + "overall": 6.3, "scores": { "silhouette": { "score": 7, - "reason": "Low pebble hull, chevron prow, chevron-nosed turret with the team roof; side stowage (tools, jerry cans) breaks the hull line.", + "reason": "Low pebble hull, chevron prow, octagonal team-red turret roof and a long gun with a muzzle brake: reads as a light tank from every view. The turret is small and flat, and the side profile is a plain slab.", "fixes": [ - "turret 5% bigger for RTS exaggeration" + "turret ~10% bigger and taller for RTS exaggeration", + "side skirts or stowage that break the long hull line" ] }, "proportions": { "score": 7, - "reason": "6.1 x 3.05 x 3.2 m, 0.6 m road wheels: inside the AFV-playbook light-tank ranges.", - "fixes": [] + "reason": "6.1 x 3.1 x 3.2 m with 0.6 m road wheels: inside the AFV-playbook light-tank range.", + "fixes": [ + "check the gun length (~0.6 of hull length) against the playbook light-tank range; it reads long" + ] }, "materials": { - "score": 7, - "reason": "steel_painted bone/slate/team (paint film, chips show steel), iron_cast fittings and wheel discs, brass hubs/brake, rubber tyres, iron_rust links, oak tool shafts. Painted chips are texture-space dashes, elongated along U.", + "score": 5, + "reason": "The bone paint is covered in dark elongated chip dashes that read as wood grain or birch bark, so the hull looks like painted wood; rubber tyres and track links read correctly.", "fixes": [ - "rounder chip shapes need finer V texels (sheet limit)" + "rounder, edge-biased chips with lower density on large plates (a steel_painted variant with smaller chip scale)" ] }, "detail": { - "score": 7, - "reason": "Glacis applique plates with a centre shut-line, weld beads on the seams, spare links on the glacis, deck hatches with bolt rows, shovel/crowbar in clamps, jerry-can rack, grille, lamps, sensor, hatch, antenna.", + "score": 6, + "reason": "Glacis spare links, weld seams, hatches, tool rack, jerry cans, grille, lamps, antenna; plates are otherwise flat, with no bolt rows visible at hero distance.", "fixes": [ - "tow cable on the rear plate" + "bolt rows on the turret roof edge and glacis", + "a tow cable on the rear plate" ] }, "wear": { - "score": 7, - "reason": "weather() per space: hull chips 13% / dust 30% / mud 50% of the 0.9 m ground band, exhaust soot streaking down the rear plate; wheels muddied in their pivots; turret and gun chipped and dusty. Measured, gated.", + "score": 6, + "reason": "Chips everywhere and a little mud on the running gear; the chip pattern is uniform so it reads as texture rather than history.", "fixes": [ - "mud heavier on the lower hull sides" + "heavier mud on the lower hull and wheels", + "soot streak down the rear plate visible from the 3/4 view" ] }, "lod": { - "score": 8, - "reason": "LOD2 382 tris: belt slab + empty wheel pivots, team roof and deck hold (21.4% share).", - "fixes": [] + "score": 6, + "reason": "FAR LOD2 replaces the running gear with a solid belt slab and the road wheels vanish, so the side outline changes noticeably at RTS zoom.", + "fixes": [ + "keep 6 road-wheel discs per side at LOD2 (running.ts) inside an open belt" + ] }, "technical": { - "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18. Gun declared `connect: hinge` on its trunnion; wheel hubs `clearOf: ['hull']` (stub axles).", - "fixes": [] + "score": 7, + "reason": "QA PASS (rebuilt today): 32.2k / 8.52k / 574 tris; 23 / 23 / 7 draws at LOD0/1 is heavy for an RTS unit.", + "fixes": [ + "merge static wheel pivots into fewer draws (one per side) since they spin together" + ] } }, "renders": { - "hero.png": "d38553706df659c6a6a374ef9b85308409f54f38533168be7fad1308b4709d9f", - "sheet.png": "24420c294e076d3149b751ea8338ad2a5f15cbb281d480ab335313f7078d79f8", - "raking.png": "3dd9e8d65ed8662e5d13acde3b9d43e1552b10e337284c0cf068768679a7a3fe", - "raking_detail.png": "af4ecf905cb3aeeeb82d1f070cf7da1dd07e496f43b5144191fdd5c7c081ac2b", - "lods.png": "d403906300397041e550d333c5e9d4f451acbc20143f82be1276edbbded30a00", - "silhouette.png": "b26919ff40113ad3e943cea4b0e190fb17e2afbd7438cd8fa4225e1c12bac65d", - "value.png": "93dc0987b491679b3ae789b7faa246ee7e96748779e577943b790d7b5c1ac491", - "detail.png": "3865ac6cc2c404777b9c2c4fbb329037c0b8d3fc226c6bf8e563ba971744650a" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.951 - }, - { - "view": "side", - "iou": 0.93 - }, - { - "view": "3/4", - "iou": 0.989 - } - ], - "focal": { - "peakPart": "gun_barrel", - "peak": [ - 500, - 219 - ], - "focalShare": 0.87, - "focalArea": 0.696, - "top": [ - { - "part": "hull", - "share": 0.325 - }, - { - "part": "hull_dress", - "share": 0.291 - }, - { - "part": "gun_barrel", - "share": 0.124 - } - ], - "declared": [ - "hull", - "gun_barrel", - "turret" - ] - }, - "rest": { - "share": 0.566, - "spread": 0.778, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.063, - "straightShare": 0.427, - "taper": 0.872, - "shape": "friendly", - "declared": "friendly" - } + "hero.png": "a25a98c688c770dd8c157e8353482ffd71ee0d2dc9a12b350a3406334f698cbc", + "sheet.png": "e823a896bfaae36ee2c26934fa17a854058982ec29e464b5ea1561b2467f90e6" } } diff --git a/recipes/rts_worker/out/rts_worker/critique.json b/recipes/rts_worker/out/rts_worker/critique.json index 981e33a6..42909d79 100644 --- a/recipes/rts_worker/out/rts_worker/critique.json +++ b/recipes/rts_worker/out/rts_worker/critique.json @@ -1,127 +1,62 @@ { "asset": "rts_worker", "reviewer": "agent", - "overall": 7.1, + "overall": 6.6, "scores": { "silhouette": { "score": 7, - "reason": "Pack + upright pick (head enlarged 30%) + round cap read 'builder' at 64 px; IoU LOD2 0.95/0.93/0.92.", + "reason": "Red hard hat with goggles, big pack with a red lid and a pick held upright: reads \"builder\" from the 3/4 views; the pose is stiff with arms straight down.", "fixes": [ - "one asymmetric shape on the pack (antenna, rolled tarp)" + "carry the pick over the shoulder so it reads in silhouette from the front" ] }, "proportions": { "score": 7, - "reason": "Chunky 6-head figure; tool handle fits the fist (grip joint); real length now 1.06 m with the longer pick.", + "reason": "1.1 x 0.8 x 1.8 m, chunky 6-head figure; pack big for the RTS read; hands small.", "fixes": [ - "slightly bigger hands for the 64 px read" + "hands ~15% bigger for the 64 px read" ] }, "materials": { - "score": 7, - "reason": "Real strips per role: canvas suit (weave relief), leather gloves/boots/belt, ceramic hard hat and pack lid, painted-steel pack, cast-iron pick head, oak handle, brass canister, rubber cable. Reads as cloth vs leather vs metal under raking light; face still plain skin.", + "score": 6, + "reason": "Olive canvas suit, brown leather gloves and boots, red ceramic hat, grey steel pack separate in hero.png; the face is a plain smooth tan surface with no features, and in sheet.png boots render pale blue-grey.", "fixes": [ - "face planes beyond nose/jaw; stitched seams on the suit" + "face planes (brow, nose, cheek) or a simple face decal", + "stitched seams on the suit (normal detail)" ] }, "detail": { - "score": 7, - "reason": "Tool belt with pouch and mallet, cable coil, canister, pockets, spare pick head, goggles, nose/jaw planes on the head.", + "score": 6, + "reason": "Tool belt, pouch, cable coil, canister, goggles, pick; no knee pads or pocket flaps visible, the suit is smooth tubes.", "fixes": [ - "knee pads, chest pocket flaps" + "knee pads and chest pocket flaps", + "a rolled tarp on the pack" ] }, "wear": { - "score": 6, - "reason": "weather(): chips on 14% of hard convex edges, salt dust on 30% of up-faces, brine mud on 50% of the ground band (boots, shins); cloth/straps dust+mud only. Measured and gated. Subtle at hero distance; LOD2 colour still carries the read.", + "score": 5, + "reason": "weather() is measured, but the figure reads clean at hero distance; boots and knees carry no visible dirt.", "fixes": [ - "stronger knee dirt; worn cap peak" + "stronger knee and boot dirt, a worn hat brim" ] }, "lod": { - "score": 8, - "reason": "LOD2 328 tris keeps the cap, pack lid and pick; team share 30.6% at 64 px.", - "fixes": [] + "score": 7, + "reason": "FAR LOD2 (312 tris) keeps hat, pack and pick; boots become grey blocks.", + "fixes": [ + "keep boot colour consistent between LODs" + ] }, "technical": { "score": 8, - "reason": "QA PASS on every gate incl. connections and weather (exact coverage). Deterministic, one trim material + team slot, 8 liveries; `bun test recipes/rts_faction` 18/18.", - "fixes": [] + "reason": "QA PASS (rebuilt today): 10.1k / 1.79k / 312 tris, 3 / 3 / 2 draws.", + "fixes": [ + "spend LOD0 triangles on the face and hands rather than the pack box edges" + ] } }, "renders": { - "hero.png": "b1d432c470bb58d04ef648aba721ad4724ee0e19a8313c63279a8a36daaf762f", - "sheet.png": "c6782269aebdf8299c73f2b20a20c691fcf4ec3bf5773a1ada4ead66e7698675", - "raking.png": "24f568466fb35a2361663eddd2236cc663d0f592b979dec44c775881c5cb7314", - "raking_detail.png": "3b1093f673b43d191910736afcc8f276e1fd159231dd4f892ffdfed848798428", - "lods.png": "e37d755d65d9ba648117c5e1b08ad1ad2519fc646777b065a49a53ee05fbfc08", - "silhouette.png": "d1d9a08497203242446fe10c53674b5fb92a185fbde1a90a2cf13acc6cba9275", - "value.png": "16e2e23009ed9a6a5332576f785e01e85ab526ed6a33a46aa417d9d1e1c32307", - "detail.png": "3d41efa8e2da48b778de64c47e1f4aab4646f7a83ee0ca5f970e793b49c9626e" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.948 - }, - { - "view": "side", - "iou": 0.928 - }, - { - "view": "3/4", - "iou": 0.916 - } - ], - "focal": { - "peakPart": "cap", - "peak": [ - 417, - 71 - ], - "focalShare": 0.368, - "focalArea": 0.04, - "top": [ - { - "part": "cap", - "share": 0.236 - }, - { - "part": "pack_lid", - "share": 0.17 - }, - { - "part": "tool_head", - "share": 0.165 - } - ], - "declared": [ - "cap", - "goggles" - ] - }, - "rest": { - "share": 0.206, - "spread": 0.539, - "min": 0.2 - }, - "grounded": { - "rise": 0.118, - "bands": [ - 0.199, - 0.24, - 0.227, - 0.257, - 0.309 - ] - }, - "shape": { - "acuteShare": 0.067, - "straightShare": 0.646, - "taper": -0.131, - "shape": "stable", - "declared": "stable" - } + "hero.png": "1edb59bc7d4c345c59178127595ce6df2b26add2309a67ca0fa7dc305096c86f", + "sheet.png": "1b8b0b83d19f2b761146715eecd870c47af4f8fefd6ce4cba33a8618afdcd3fe" } } diff --git a/recipes/signal_gantry/out/signal_gantry/critique.json b/recipes/signal_gantry/out/signal_gantry/critique.json index 7db00f8c..eab4b805 100644 --- a/recipes/signal_gantry/out/signal_gantry/critique.json +++ b/recipes/signal_gantry/out/signal_gantry/critique.json @@ -1,35 +1,41 @@ { "asset": "signal_gantry", "reviewer": "agent", - "overall": 7.3, + "overall": 7, "scores": { "silhouette": { "score": 8, - "reason": "Lattice signal gantry with timber walkway and handrail, latticed posts on concrete footings, knee braces, two (or four) dolls with lower-quadrant semaphore arms, spectacles and lamps. Reads instantly.", - "fixes": [] + "reason": "Lattice signal gantry with timber walkway and handrail, latticed posts on footings, knee braces, two dolls with lower-quadrant semaphore arms: reads instantly from the front and 3/4 views.", + "fixes": [ + "ladder up one post so the access reads in silhouette" + ] }, "proportions": { "score": 8, - "reason": "Span, post height and arm sizes pass the dossier measures; walkway at a believable working height.", - "fixes": [] + "reason": "1.1 x 9.9 x 13.1 m; span, post height and arm sizes pass the dossier measures; walkway at a believable working height.", + "fixes": [ + "check the doll height above the walkway against the dossier photos (they read slightly tall)" + ] }, "materials": { - "score": 7, - "reason": "Painted steel lattice (chips show steel), rusted dark fittings, weathered timber walkway, painted red/white arms on wood_painted, glowing lenses, concrete footings. Black paint reads near-uniform at distance.", + "score": 6, + "reason": "Black painted lattice, timber walkway, red arms, lamps; the black paint is near-uniform with no value variation, so the lattice reads as one flat black mass.", "fixes": [ - "slightly lighter, sun-faded top surfaces on the lattice" + "sun-faded lighter tops on the lattice members and a slightly lighter painted black", + "galvanised grey handrail" ] }, "detail": { "score": 7, - "reason": "Lattice webs, battens, knee braces, ladder, handrails, finials, spectacle glasses, lamp cases, balance weights.", + "reason": "Lattice webs, battens, knee braces, handrails, finials, spectacles, lamp cases, balance weights.", "fixes": [ - "operating rods/wires down the posts to the ground" + "operating rods/wires down the posts to the ground", + "rivet heads at lattice nodes" ] }, "wear": { - "score": 6, - "reason": "Ground splash on the footings and post feet, dust on the walkway, chipped paint from the strip. No rust streaks below rivets, no soot above the tracks.", + "score": 5, + "reason": "Some splash on the footings; otherwise clean at hero distance: no rust streaks, no soot above the tracks.", "fixes": [ "soot on the girder underside above each track", "rust streaks below the lattice nodes" @@ -37,76 +43,21 @@ }, "lod": { "score": 7, - "reason": "LOD2 keeps the frame, walkway, dolls and arms; the latticed posts become solid black boxes (noticeably heavier at distance).", + "reason": "FAR LOD2 keeps the frame, walkway, dolls and arms; the latticed posts become solid black boxes, noticeably heavier at distance.", "fixes": [ "LOD2 posts as open frames (four angles + three battens) instead of solid boxes" ] }, "technical": { "score": 8, - "reason": "QA PASS both variants: 8 376 / 3 684 / 800 and 10 876 / 4 944 / 1 404; arms pivot as moving parts. Now 8 456 / 3 748 / 800. Connections gate clean (was 2 findings): the ball finial sits on a collar sleeved over the mast's open top (LOD0/1; LOD2 keeps the plain ball inside its budget).", - "fixes": [] + "reason": "QA PASS (rebuilt today): 9.1k / 3.98k / 768 tris, 6 / 6 / 6 draws (arms pivot as moving parts).", + "fixes": [ + "merge the static lattice and walkway into one draw" + ] } }, "renders": { - "hero.png": "fb267a32cfd436fc6281a05c3b8c401a219f304d7f1c8cbf4019dec1f53344c2", - "sheet.png": "68af4af7f52d04ce0ce5a43641709acb9830b22e3fe60069dd2855693626a208", - "raking.png": "b659d04d644b6f6c24763f260d922ea696d932e991ad01d3568e49549a5205af", - "raking_detail.png": "86f68ab73c56844b8739bc7f4e1a38de795fb7feb9cffdb4e1835e4ae36a7ca3", - "lods.png": "7c76ea4440bb7f20a3df8ef3a8099ec6c8a91174cd3023803e3ef2cff976fc4d", - "silhouette.png": "d79868a1934c4bf40f5a6502ac48a0bb2bec04398ab8d88bf0c1bb5236cf7667", - "value.png": "2438c69120409eb2b1528523e5ae8fe74e3192ad23f30bc6bffc5b12bee55d41", - "detail.png": "c0120f7be6a61298e2d80ae6a746fdb95f6f7b32916b2d01f8c9602444686978" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.944 - }, - { - "view": "side", - "iou": 0.903 - }, - { - "view": "3/4", - "iou": 0.871 - } - ], - "focal": { - "peakPart": "walkway", - "peak": [ - 318, - 191 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walkway", - "share": 0.341 - }, - { - "part": "girder", - "share": 0.283 - }, - { - "part": "post_r", - "share": 0.165 - } - ], - "declared": [] - }, - "rest": { - "share": 0.137, - "spread": 0.605, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.263, - "straightShare": 0.856, - "taper": 0.42, - "shape": "stable" - } + "hero.png": "3755432d213bf7c8518297a7849fe321589ad23cb910ae53da1295f2aec8ad93", + "sheet.png": "50ca482ba82f13b8c6961c6b303552db00d601a2995aa2619e3d37895dd3776b" } } diff --git a/recipes/signal_gantry/out/signal_gantry_4/critique.json b/recipes/signal_gantry/out/signal_gantry_4/critique.json index 08946346..f6796650 100644 --- a/recipes/signal_gantry/out/signal_gantry_4/critique.json +++ b/recipes/signal_gantry/out/signal_gantry_4/critique.json @@ -1,35 +1,41 @@ { "asset": "signal_gantry_4", "reviewer": "agent", - "overall": 7.3, + "overall": 7, "scores": { "silhouette": { "score": 8, - "reason": "Lattice signal gantry with timber walkway and handrail, latticed posts on concrete footings, knee braces, two (or four) dolls with lower-quadrant semaphore arms, spectacles and lamps. Reads instantly.", - "fixes": [] + "reason": "Four-doll version of the lattice gantry spanning ~18 m: reads instantly; four dolls evenly spaced along the walkway.", + "fixes": [ + "ladder up one post so the access reads in silhouette" + ] }, "proportions": { "score": 8, - "reason": "Span, post height and arm sizes pass the dossier measures; walkway at a believable working height.", - "fixes": [] + "reason": "1.1 x 17.9 x 13.1 m; span and post height pass the dossier measures.", + "fixes": [ + "a central support or deeper girder: an 18 m span with this girder depth reads light against real four-track gantries" + ] }, "materials": { - "score": 7, - "reason": "Painted steel lattice (chips show steel), rusted dark fittings, weathered timber walkway, painted red/white arms on wood_painted, glowing lenses, concrete footings. Black paint reads near-uniform at distance.", + "score": 6, + "reason": "Same as the two-doll gantry: near-uniform black lattice, timber walkway, red arms.", "fixes": [ - "slightly lighter, sun-faded top surfaces on the lattice" + "sun-faded lighter tops on the lattice members", + "galvanised grey handrail" ] }, "detail": { "score": 7, - "reason": "Lattice webs, battens, knee braces, ladder, handrails, finials, spectacle glasses, lamp cases, balance weights.", + "reason": "Lattice webs, battens, knee braces, handrails, four dolls with spectacles and lamps.", "fixes": [ - "operating rods/wires down the posts to the ground" + "operating rods/wires down the posts to the ground", + "rivet heads at lattice nodes" ] }, "wear": { - "score": 6, - "reason": "Ground splash on the footings and post feet, dust on the walkway, chipped paint from the strip. No rust streaks below rivets, no soot above the tracks.", + "score": 5, + "reason": "Clean at hero distance; no rust streaks or soot.", "fixes": [ "soot on the girder underside above each track", "rust streaks below the lattice nodes" @@ -37,76 +43,21 @@ }, "lod": { "score": 7, - "reason": "LOD2 keeps the frame, walkway, dolls and arms; the latticed posts become solid black boxes (noticeably heavier at distance).", + "reason": "FAR LOD2 keeps frame, walkway, dolls and arms; latticed posts become solid black boxes.", "fixes": [ - "LOD2 posts as open frames (four angles + three battens) instead of solid boxes" + "LOD2 posts as open frames instead of solid boxes" ] }, "technical": { "score": 8, - "reason": "QA PASS both variants: 8 376 / 3 684 / 800 and 10 876 / 4 944 / 1 404; arms pivot as moving parts. Now 11 036 / 5 072 / 1 404. Connections gate clean (was 4 findings): finial collars sleeved over each mast top (LOD0/1).", - "fixes": [] + "reason": "QA PASS (rebuilt today): 12.3k / 5.54k / 1.34k tris, 10 / 10 / 10 draws (arms pivot).", + "fixes": [ + "merge the static lattice and walkway into one draw" + ] } }, "renders": { - "hero.png": "dfe0cb439627a5c085e949508630735624be37bb9bebfe492cb79914ad161b2e", - "sheet.png": "650764db407ca5461eadf58a8d29593f38cf79e0fb358cf3e72182251f409a34", - "raking.png": "88cd84dbfd32c205a0d8e4a05cb5b212fb46252ba1f7756e63d0145b63c7cf89", - "raking_detail.png": "a53003462467af317e8effcc50f8db4d8134905fcf01b40cd230558a469557a9", - "lods.png": "52238c1b3063ed44ffb085e4a67432b7a1a439c35239eb919c4bbbad4aabc470", - "silhouette.png": "8a8929e794ac7c255a7042f0fa7c01649defd09acdf52a3cb0f8b51eaa06eb7b", - "value.png": "c6f0ac03b5c10eaee3bb854bb91e5bb41ba7334d03d5ee4c25b12759e0b5c042", - "detail.png": "3ab9e0833df241d29bc20d55281da79b36631cd19b1435f8354e082a2b0b9f4a" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.727 - }, - { - "view": "side", - "iou": 0.903 - }, - { - "view": "3/4", - "iou": 0.85 - } - ], - "focal": { - "peakPart": "walkway", - "peak": [ - 252, - 196 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "walkway", - "share": 0.403 - }, - { - "part": "girder", - "share": 0.318 - }, - { - "part": "post_r", - "share": 0.118 - } - ], - "declared": [] - }, - "rest": { - "share": 0.173, - "spread": 0.617, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.376, - "straightShare": 0.823, - "taper": 0.018, - "shape": "stable" - } + "hero.png": "0e50c3fa572c910c8a471f4c1b6a1525b3646b1efce93dc415a228ef634dfa9e", + "sheet.png": "fbe1910da292e448155b448983320a20a9f7ad6f52aa8ab57e90012417c757f5" } } diff --git a/recipes/sound_demo/out/sound_demo/critique.json b/recipes/sound_demo/out/sound_demo/critique.json new file mode 100644 index 00000000..d4a7e71d --- /dev/null +++ b/recipes/sound_demo/out/sound_demo/critique.json @@ -0,0 +1,121 @@ +{ + "asset": "sound_demo", + "reviewer": "agent", + "overall": 6, + "scores": { + "silhouette": { + "score": 7, + "reason": "Door in a frame, bell on a yoke, crate and a V10 block on a deck: each exhibit reads as itself from the front and 3/4 views.", + "fixes": [ + "a visible clapper and headstock on the bell so it reads as a swinging bell" + ] + }, + "proportions": { + "score": 7, + "reason": "Deck 1.2 x 3.4 x 1.8 m; door 0.6 x 1.6 m opening, 300 mm bell and 0.8 m crate match the README table; the bell yoke is tall and thin for a 300 mm bell.", + "fixes": [ + "shorter, sturdier yoke posts (~1.2 m) with a cross-tie" + ] + }, + "materials": { + "score": 5, + "reason": "The door leaf shows blotchy darker rectangles that read as patching, the frame is white with heavy dark scratches, the deck is dark grey with scratches; crate pine and engine iron read. In sheet.png the crate renders flat white.", + "fixes": [ + "clean painted pine on the door leaf with panel mouldings instead of rectangle patches", + "cast-iron engine block lighter in value so it separates from the black heads" + ] + }, + "detail": { + "score": 5, + "reason": "Door with handle and hinges, bell, crate with battens and corner irons, engine block with heads and pulleys; no manifolds, plug leads or bell clapper.", + "fixes": [ + "intake and exhaust manifolds on the V10", + "bell clapper and headstock" + ] + }, + "wear": { + "score": 4, + "reason": "Scratches on the frame and deck read as a texture pattern, not use; nothing has a history (no oil under the engine, no worn door edge).", + "fixes": [ + "an oil stain under the engine bearers", + "worn paint at the door handle and lock edge" + ] + }, + "lod": { + "score": 6, + "reason": "lods.png: LOD2 crate loses its frame and battens to a plain box and the door loses its panels; outline holds.", + "fixes": [ + "keep the crate frame as 4 edge strips at LOD2", + "keep the door panel mouldings as vertex colour at LOD2" + ] + }, + "technical": { + "score": 8, + "reason": "QA clean (pack): 5.3k / 2.9k / 1.1k tris; audio emitters exported.", + "fixes": [ + "LOD2 at 1.1k is high for four small props: drop the engine pulleys and bell yoke detail at LOD2" + ] + } + }, + "renders": { + "hero.png": "5f9c74941a39d75476dd7fb4f5d2f78e7eb4ab5933277198d214dc5510b3b09e", + "sheet.png": "4df212916b772de55768d42fc7d3836a90df84da5c33513af1d101c36aa058ea", + "raking.png": "7a597f29a6eca751a74419b862cd195adda30d2885d287e9cb3bf58c0e759331", + "raking_detail.png": "2ed76b5cdb451b4899a02917d6289b728820bdbfb227a1040ad4012d9ec2f6da", + "lods.png": "7b0224626300765f5466170aa392e94569da5f8d7bf3a93bfe66c3e0050bc5b6", + "silhouette.png": "aebdd13b52466ac1629e7139f70c671ef6127adc8e84006ac4499fa664df0765", + "value.png": "e9196e5b30f58b3c39e0e1badec49e542e5867b0973017f83a1070433ac53a4d", + "detail.png": "9957caf3d16a8b2ec4bc6df22ac40f528d0e0bf3b293905434341d556825cd89" + }, + "metrics": { + "silhouette": [ + { + "view": "front", + "iou": 0.996 + }, + { + "view": "side", + "iou": 0.997 + }, + { + "view": "3/4", + "iou": 0.992 + } + ], + "focal": { + "peakPart": "door_frame", + "peak": [ + 283, + 62 + ], + "focalShare": 0, + "focalArea": 0, + "top": [ + { + "part": "door_frame", + "share": 0.614 + }, + { + "part": "door_leaf", + "share": 0.36 + }, + { + "part": "deck", + "share": 0.026 + } + ], + "declared": [] + }, + "rest": { + "share": 0.642, + "spread": 0.748, + "min": 0.25 + }, + "shape": { + "acuteShare": 0.026, + "straightShare": 0.88, + "taper": 0.73, + "shape": "stable" + } + } +} diff --git a/recipes/station_halt/out/station_halt/critique.json b/recipes/station_halt/out/station_halt/critique.json index 631d2aaf..e62170f6 100644 --- a/recipes/station_halt/out/station_halt/critique.json +++ b/recipes/station_halt/out/station_halt/critique.json @@ -1,47 +1,50 @@ { "asset": "station_halt", "reviewer": "agent", - "overall": 7.3, + "overall": 7, "scores": { "silhouette": { "score": 8, - "reason": "GWR country station (Toddington): brick building with slate roof and chimneys, long valanced platform canopy on cast-iron columns, lamps, name board, barrow and churns on the platform. Reads from every view.", + "reason": "GWR country station: brick building with slate roof and two chimneys, long valanced platform canopy on cast-iron columns, lamps, name board, barrow and churns. Reads from every view.", "fixes": [ - "fretted canopy valance profile at LOD1" + "fretted canopy valance profile visible at LOD1" ] }, "proportions": { "score": 8, - "reason": "Overall box and the dossier measures pass; door, window and canopy heights read at human scale against the 1.8 m figure.", - "fixes": [] + "reason": "11.6 x 30.1 x 8.3 m; door, window and canopy heights read at human scale; dossier measures pass.", + "fixes": [ + "check chimney stack height against the Toddington photos (they read short)" + ] }, "materials": { - "score": 7, - "reason": "Chocolate-and-cream timber now wood_painted (grain and chips through the paint), cast-iron columns/brackets as iron_cast, brick, slate, stone, glass from the classic strips. Brick and slate are still the greyscale classic strips tinted, flatter than the library strips.", + "score": 6, + "reason": "Red brick, cream valance, chocolate columns and stone platform read; the slate roof and canopy read as flat grey sheets with no slate courses at hero distance, and the brick is a uniform red without course variation.", "fixes": [ - "blue-brick plinth and red brick as distinct project materials (brick colour variation per course)", - "slate with materialVariant per roof slope" + "slate courses with per-slate value jitter on the roof", + "brick colour variation per course and a blue-brick plinth" ] }, "detail": { "score": 7, - "reason": "Recessed sash windows, doors, chimney pots, ridge/coping, canopy valance, columns with brackets, lamps with glass, name board, room boards, clock, rainwater goods, benches, barrow, churns.", + "reason": "Sash windows, doors, chimney pots, canopy valance, columns with brackets, lamps, name board, benches, barrow, churns.", "fixes": [ "platform edge coping slabs with joints", - "gutter brackets" + "gutter brackets and downpipes on the gable" ] }, "wear": { - "score": 6, - "reason": "Ground splash along the platform and building base, dust on top faces, AO. No soot on the canopy underside, no moss on the slates or green algae on the north wall, paint not faded.", + "score": 5, + "reason": "Clean at hero distance: no soot on the canopy underside, no moss on slates, no staining under the windows.", "fixes": [ - "soot streaks on the canopy underside from locomotives", - "moss/lichen on north-facing slates and the platform edge" + "soot streaks on the canopy underside", + "moss/lichen on north slates and the platform edge", + "damp at the brick plinth" ] }, "lod": { "score": 7, - "reason": "LOD2 (460) keeps building, roof, chimneys, canopy, columns, lamps and the board shape; the paved platform turns flat light grey and the name board loses its text.", + "reason": "FAR LOD2 (460) keeps building, roof, chimneys, canopy, columns and the board; the platform turns flat light grey and the board loses its lettering.", "fixes": [ "keep the platform strip at LOD2 so its value matches", "keep the name-board lettering as one quad strip" @@ -49,69 +52,14 @@ }, "technical": { "score": 8, - "reason": "QA PASS, 8 770 / 4 850 / 460, two draws (opaque + glass). Now 8 802 / 4 882 / 460. Connections gate clean (was 1 finding): the eaves band is four closed brick courses bearing on the walls (inner face 2 mm off, 3 cm proud) instead of an open ring hovering 3 cm out.", - "fixes": [] + "reason": "QA PASS (rebuilt today): 8.56k / 4.74k / 460 tris, 2 / 2 / 2 draws: lean for a building this size.", + "fixes": [ + "spend a few hundred more LOD0 triangles on the canopy valance fret" + ] } }, "renders": { - "hero.png": "aff0e5d3d5554a8fe1f0678731c946bf7b4db9187107236ad490ed0f0fdce82b", - "sheet.png": "9aa6273894cc9cf331356b625acc2282027e4b11397d3c221ce97e729b559cf8", - "raking.png": "35b400b2dbde20d9d3ab5889148f34bc3fb41e6821d24aef95eef966a3119085", - "raking_detail.png": "3ff0486a04445022513874b2f77fa1047228a648804339652f8a9f97886b43be", - "lods.png": "5bd1616e4cb1c2164c520454cf90712364ba76435246127f69fd74d0ebf4b581", - "silhouette.png": "a7cd756bdac9e58254293dc3cdcbc844d74dfba7c8314d8e20e224cda804e7cc", - "value.png": "6d09a2b01d6459c50c1ffd6b131a86775b078ed9fb12671edb70756ba379e0e6", - "detail.png": "1891f6dad10b99c8a01444d2e1b98ee71bf1f1f0ab81d46a64c2a6feb6d633a8" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.949 - }, - { - "view": "side", - "iou": 0.96 - }, - { - "view": "3/4", - "iou": 0.97 - } - ], - "focal": { - "peakPart": "platform", - "peak": [ - 483, - 272 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "platform", - "share": 0.968 - }, - { - "part": "roof", - "share": 0.015 - }, - { - "part": "lamp_1_glass", - "share": 0.007 - } - ], - "declared": [] - }, - "rest": { - "share": 0.753, - "spread": 0.739, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.3, - "straightShare": 0.759, - "taper": 0.643, - "shape": "stable" - } + "hero.png": "0f1693cbf54f5743709f2e6d0f794c870c9a7b99c8496f70c3fc87ac169bfc39", + "sheet.png": "0601775cdeba7521f39999ec5e93a93150e0a9e917e56ee2bc252f7159872231" } } diff --git a/recipes/svg_showcase/out/svg_showcase/critique.json b/recipes/svg_showcase/out/svg_showcase/critique.json new file mode 100644 index 00000000..ec74778c --- /dev/null +++ b/recipes/svg_showcase/out/svg_showcase/critique.json @@ -0,0 +1,120 @@ +{ + "asset": "svg_showcase", + "reviewer": "agent", + "overall": 5.7, + "scores": { + "silhouette": { + "score": 7, + "reason": "Gear, amphora and goblin head on a plinth: each reads from front and 3/4 views; the star emblem is a speck on the plinth edge.", + "fixes": [ + "a bigger star emblem (0.10 m) so it reads in the hero" + ] + }, + "proportions": { + "score": 7, + "reason": "Design spec, desk-diorama scale: 0.5 x 1.8 x 0.6 m; objects match the README sizes and sit well on the plinth.", + "fixes": [ + "set the head on a short neck stump so it does not sit chin-down on the plinth" + ] + }, + "materials": { + "score": 4, + "reason": "Flat untextured colours: the gear is grey plastic, the amphora smooth orange, the head smooth green; nothing reads as metal, fired clay or skin.", + "fixes": [ + "gear on a steel strip with metalness and machined roughness", + "amphora on a terracotta strip with throw rings", + "skin strip with roughness variation on the head" + ] + }, + "detail": { + "score": 5, + "reason": "Gear teeth with chamfers, lightening holes, amphora rim and foot, sculpted eyes, brow, nostrils and tusks; surfaces between are bare.", + "fixes": [ + "bevelled gear bore with a keyway", + "skin wrinkles and ear notches on the head" + ] + }, + "wear": { + "score": 3, + "reason": "None: everything is pristine and uniform.", + "fixes": [ + "grime() oil on the gear teeth and dust on the amphora shoulder" + ] + }, + "lod": { + "score": 7, + "reason": "FAR LOD2 keeps all three objects, outlines and colours; the head becomes faceted.", + "fixes": [ + "a slightly finer sculpt cell on the head at LOD2" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean (pack): 38.7k / 10.2k / 1.9k tris; most of LOD0 goes to the sculpted head on a desk prop.", + "fixes": [ + "coarser sculpt cell on the head at LOD0 (the smooth areas do not need it)" + ] + } + }, + "renders": { + "hero.png": "54578ff672cff6d71421adc6f0846bb8bb411a6e47d387a36814e2dfbd31ea29", + "sheet.png": "c8c6c5294dfae696b112e40d81b9a55882d82c7d4f80a7bb0913962676f534b0", + "raking.png": "359a4dac93250d6df93c773f6ed2bdefed8f13eb9a27b058dc5ae6976a58ab09", + "raking_detail.png": "2bd9c3f7d1eb1bd1f51f7d56116662e6d4ab8ebbab5292e9899a5d6e88abf35c", + "lods.png": "4f3e7a47b8f0d2a9d4abe07715f3a5be060091583657a77ea44fff72e29f4010", + "silhouette.png": "577471fe7e48930351616208623c2bbdd531b9e1f154de6aa78863f73e8b50ae", + "value.png": "a297ab94a4c5693831110c0bd1ebf445b35c0c526af5e427c593865e68839ba5", + "detail.png": "8aa9ca3e8339b7a695293a719f34d5013f35683e519e080482ab21f044598418" + }, + "metrics": { + "silhouette": [ + { + "view": "front", + "iou": 0.978 + }, + { + "view": "side", + "iou": 0.956 + }, + { + "view": "3/4", + "iou": 0.982 + } + ], + "focal": { + "peakPart": "head", + "peak": [ + 620, + 141 + ], + "focalShare": 0, + "focalArea": 0, + "top": [ + { + "part": "gear", + "share": 0.568 + }, + { + "part": "head", + "share": 0.246 + }, + { + "part": "amphora", + "share": 0.108 + } + ], + "declared": [] + }, + "rest": { + "share": 0.75, + "spread": 0.663, + "min": 0.25 + }, + "shape": { + "acuteShare": 0.13, + "straightShare": 0.593, + "taper": 0.568, + "shape": "stable" + } + } +} diff --git a/recipes/t34/out/t34/critique.json b/recipes/t34/out/t34/critique.json index 38fbdd1a..7379ef8e 100644 --- a/recipes/t34/out/t34/critique.json +++ b/recipes/t34/out/t34/critique.json @@ -1,71 +1,72 @@ { "asset": "t34", "reviewer": "agent", - "overall": 7.4, + "overall": 6.7, "scores": { "silhouette": { "score": 8, - "reason": "Unmistakable T-34/76: 60\u00b0 glacis, 40\u00b0 side plates, five big Christie wheels per side, rear sprocket, hex 'nut' turret with cupola (1941 variant: cast turret, big hatch), side fuel drums, F-34 gun. Reads from all six views and from LOD2.", + "reason": "Unmistakable T-34: sloped glacis and side plates, five Christie road wheels plus idler and sprocket, hex turret with twin cupola hatches, rear fuel drums, F-34 gun. Reads from all six views and LOD2.", "fixes": [ - "round the cast-turret variant's edges (cast radius) rather than faceted rings" + "soften the turret edges (cast radius) so it is not a sharp-edged box" ] }, "proportions": { "score": 8, - "reason": "Overall 6.7 \u00d7 3.0 \u00d7 2.45 m against the 6.68 \u00d7 3.00 \u00d7 2.45 m spec; wheel 830 mm, track 550 mm, track centres, hull 5.92 m and barrel 3.162 m pass the \u00b12% measures. Tracks rest on z = 0 at every LOD (test added).", + "reason": "6.8 x 3.1 x 2.5 m against the 6.68 x 3.00 x 2.45 m spec; wheel 830 mm, track 550 mm and barrel pass the measures.", "fixes": [ - "check turret ring position against a side elevation drawing (turret slightly far forward by eye)" + "check the turret ring position against a side elevation (turret reads slightly far forward)" ] }, "materials": { - "score": 7, - "reason": "4BO green as steel_painted with per-tank jitter, rubber tyres, rusted manganese track, cast-iron fittings, rusted exhausts, all at true scale with per-triangle box projection. Chips are uniform dark dots rather than edge-led; spare track links on the glacis read rust-red, which is right.", + "score": 5, + "reason": "Green paint reads, rubber tyres read, spare links read rust-red; but the track links read as brown wooden slats under raking light, and the chips are uniform black dots spread over every plate like spatter.", "fixes": [ - "edge-biased chipping instead of the strip's uniform chip pattern", - "slightly glossier worn paint on hatch rims" + "track links in a dark steel/iron_rust with a lighter worn grouser face instead of the brown wood-like strip", + "edge-biased chipping instead of uniform dots" ] }, "detail": { "score": 7, - "reason": "Link-by-link track with guide horns, dished Christie wheels with ribs and bolts, grab rails, glacis spare links, drivers' hatch with periscopes, MG ball, fuel drums with straps, exhausts, slogan. Missing weld beads, tow cable, tool boxes; LOD0 at 45.7k is well spent.", + "reason": "Link track with horns, dished wheels with ribs, grab rails, glacis spare links, hatches, MG ball, fuel drums with straps, exhausts, slogan.", "fixes": [ - "weld beads at the glacis/side-plate joints", - "tow cable along the hull side with eyes", + "weld beads at plate joints", + "tow cable along the hull side", "tool box on the fender" ] }, "wear": { - "score": 7, - "reason": "Ragged mud on tracks, wheels and lower hull, dust lift on the deck and turret roof, exhaust soot on the rear plate, soot-stained muzzle, chipped paint. Paint is still uniform across panels; no scratches or rust streaks.", + "score": 4, + "reason": "Only the dot chips: in hero.png and raking.png the wheels, tracks and lower hull carry no visible mud or dust, the exhausts no soot on the rear plate.", "fixes": [ - "rust streaks below the exhausts and grab rails", - "scuffed bare steel where crew climb onto the glacis" + "grime({ ground }) mud on tracks, wheels and lower hull", + "exhaust soot on the rear plate", + "dust on the deck" ] }, "lod": { "score": 7, - "reason": "LOD2 (1.6k) now keeps the five Christie wheels per side as discs inside an open belt instead of a solid slab; outline and colour match LOD0; slogan and fuel drums drop.", + "reason": "lods.png: LOD2 (1.7k) keeps the outline, turret, wheels as discs in an open belt and colour; spare links and slogan drop.", "fixes": [ - "keep the fuel drums and slogan at LOD2 (a handful of triangles)" + "keep the glacis spare links as two dark quads at LOD2" ] }, "technical": { "score": 8, - "reason": "QA PASS both variants: 45 750 / 11 084 / 1 588 (t34), 44 124 / 10 640 / 1 542 (t34_1941), one material, rests on the ground at every LOD (assertUprightOnGround test). Now 49 502 / 11 092 / 1 596. Connections gate clean (was 23 findings per variant): track links are a pinned chain (hinge pin at each link's leading edge, ~8 mm into both links), front mudguards run a 15 cm tab under the pannier floor, headlight sits on the glacis, turret grab-handle feet run into the side plate.", + "reason": "QA clean (pack): 48.2k / 10.4k / 1.7k tris, one material.", "fixes": [ - "declare `real.mass` (26.5 t) and a physics shell so the mass gate covers it" + "declare real.mass (26.5 t) and a physics shell so the mass gate covers it" ] } }, "renders": { - "hero.png": "849fe12bfade2cd2e290b1595837f3fc2fee75bd56a69288ec5c1a87ee736898", - "sheet.png": "f5ebf51253b09a830796170b08827038a1264857d0654f6e96dd25271e21a421", - "raking.png": "de9fffbb345e9c199b288a5e75323f7a4a306337799d9310e65455adff8a0e8c", - "raking_detail.png": "ecd99ff15a1cf883e49075a58fe92f672d14dd002621ffef5501214be02ba147", - "lods.png": "d360c46f471292eb9c6156565d9be55d73c7442917387e8b94aa946f251ac848", - "silhouette.png": "7332b23d121d786462ede866c35fa3ffc70018e380e54a2b12f5e0bfa0b26658", - "value.png": "96b27bd2392191786aaa00207430fed4adeefd7df26dcb1a71fb3a818c37ede4", - "detail.png": "efa5c8b341f65f70c5f5898b128d236fac3558b3358c35141b4510b4f3c1f728" + "hero.png": "fe265cbf198fc8d4f65057601b0622a2f96b9daeed7400f94f7b26fa1d4de177", + "sheet.png": "09459a8b40af00dd37aac94207d4a1b374d8dca8fdf6a4185968a89394abb2e5", + "raking.png": "5220ed096a7211add507eb057f74023902176081e3f4309d75ac053afc6e083d", + "raking_detail.png": "83af90b2cb9708c8b140348800fb85dcb64d22b72e49538c1fe8a3bca31c13ad", + "lods.png": "b773835030ec6b29aa7ded4d6a41a185b932ce274a80d252e16a6221cb281a9d", + "silhouette.png": "d063d4af0aced926df68cc493537ab3cfdd26255e69f7a855a4ab3e6caf7ce72", + "value.png": "61365a41474dd50b246aebc281be6efefff430db18cc002f0749ca6f76cbf97f", + "detail.png": "4c8acd18fda3f243e9572d984fa80f65cecb629756f8b014e2953fb5fb7d750e" }, "metrics": { "silhouette": [ @@ -75,7 +76,7 @@ }, { "view": "side", - "iou": 0.9 + "iou": 0.906 }, { "view": "3/4", @@ -86,29 +87,29 @@ "peakPart": "turret_body", "peak": [ 397, - 77 + 76 ], "focalShare": 0, "focalArea": 0, "top": [ { "part": "turret_body", - "share": 0.616 + "share": 0.613 }, { "part": "fittings", - "share": 0.259 + "share": 0.262 }, { "part": "hull", - "share": 0.068 + "share": 0.063 } ], "declared": [] }, "rest": { - "share": 0.464, - "spread": 0.817, + "share": 0.465, + "spread": 0.819, "min": 0.25 }, "shape": { diff --git a/recipes/t34/out/t34_1941/critique.json b/recipes/t34/out/t34_1941/critique.json index 990c8a22..249300da 100644 --- a/recipes/t34/out/t34_1941/critique.json +++ b/recipes/t34/out/t34_1941/critique.json @@ -1,71 +1,70 @@ { "asset": "t34_1941", "reviewer": "agent", - "overall": 7.4, + "overall": 6.7, "scores": { "silhouette": { "score": 8, - "reason": "Unmistakable T-34/76: 60\u00b0 glacis, 40\u00b0 side plates, five big Christie wheels per side, rear sprocket, hex 'nut' turret with cupola (1941 variant: cast turret, big hatch), side fuel drums, F-34 gun. Reads from all six views and from LOD2.", + "reason": "Early T-34: low flat-sided turret with a single big hatch, sloped hull, five road wheels with idler and sprocket, F-34 gun. Reads as a T-34 from every view.", "fixes": [ - "round the cast-turret variant's edges (cast radius) rather than faceted rings" + "soften the turret edges and nose (cast radius) for the 1941 turret" ] }, "proportions": { "score": 8, - "reason": "Overall 6.7 \u00d7 3.0 \u00d7 2.45 m against the 6.68 \u00d7 3.00 \u00d7 2.45 m spec; wheel 830 mm, track 550 mm, track centres, hull 5.92 m and barrel 3.162 m pass the \u00b12% measures. Tracks rest on z = 0 at every LOD (test added).", + "reason": "6.8 x 3.0 x 2.3 m against spec; measures pass.", "fixes": [ - "check turret ring position against a side elevation drawing (turret slightly far forward by eye)" + "check the turret ring position against a side elevation" ] }, "materials": { - "score": 7, - "reason": "4BO green as steel_painted with per-tank jitter, rubber tyres, rusted manganese track, cast-iron fittings, rusted exhausts, all at true scale with per-triangle box projection. Chips are uniform dark dots rather than edge-led; spare track links on the glacis read rust-red, which is right.", + "score": 5, + "reason": "As on t34: wooden-looking brown track slats and uniform black chip dots over every plate.", "fixes": [ - "edge-biased chipping instead of the strip's uniform chip pattern", - "slightly glossier worn paint on hatch rims" + "dark steel track links with worn grouser faces", + "edge-biased chipping instead of uniform dots" ] }, "detail": { "score": 7, - "reason": "Link-by-link track with guide horns, dished Christie wheels with ribs and bolts, grab rails, glacis spare links, drivers' hatch with periscopes, MG ball, fuel drums with straps, exhausts, slogan. Missing weld beads, tow cable, tool boxes; LOD0 at 45.7k is well spent.", + "reason": "Link track, dished wheels, grab rails, glacis spare links, hatches, MG ball, exhausts.", "fixes": [ - "weld beads at the glacis/side-plate joints", - "tow cable along the hull side with eyes", - "tool box on the fender" + "weld beads at plate joints", + "tow cable along the hull side" ] }, "wear": { - "score": 7, - "reason": "Ragged mud on tracks, wheels and lower hull, dust lift on the deck and turret roof, exhaust soot on the rear plate, soot-stained muzzle, chipped paint. Paint is still uniform across panels; no scratches or rust streaks.", + "score": 4, + "reason": "Only the dot chips; no visible mud on the running gear and lower hull, no exhaust soot.", "fixes": [ - "rust streaks below the exhausts and grab rails", - "scuffed bare steel where crew climb onto the glacis" + "grime({ ground }) mud on tracks, wheels and lower hull", + "exhaust soot on the rear plate" ] }, "lod": { "score": 7, - "reason": "LOD2 (1.6k) now keeps the five Christie wheels per side as discs inside an open belt instead of a solid slab; outline and colour match LOD0; slogan and fuel drums drop.", + "reason": "LOD2 keeps the outline, turret, wheels in an open belt and colour.", "fixes": [ - "keep the fuel drums and slogan at LOD2 (a handful of triangles)" + "keep the glacis spare links as two dark quads at LOD2" ] }, "technical": { "score": 8, - "reason": "QA PASS both variants: 45 750 / 11 084 / 1 588 (t34), 44 124 / 10 640 / 1 542 (t34_1941), one material, rests on the ground at every LOD (assertUprightOnGround test). Now 47 876 / 10 648 / 1 550. Connections gate clean (was 23 findings): pinned track chain, mudguard tabs under the pannier, headlight on the glacis, grab-handle feet into the turret plate.", + "reason": "QA clean (pack): 46.6k / 10.0k / 1.7k tris, one material.", "fixes": [ - "declare `real.mass` (26.5 t) and a physics shell so the mass gate covers it" + "declare real.mass (26.5 t) and a physics shell" ] } }, "renders": { - "hero.png": "457f0754e6e086d73daedb658162df390833858a1d776ad433bb79c893c5c2cf", - "sheet.png": "9bc82f624abc37aa70c137e0061666b61de5f31fc3610cb57c06063517264ce9", - "raking.png": "897b4f5ccbf023ac1d4ab5b28057ed99aea12af0e05b94bc4464eeb05f6298a7", - "raking_detail.png": "138bffdf871d765241f7a9316547a5571620fa066df77a50c94b32940e3a6e30", - "lods.png": "5496a700c22fbb7119b51f692ac783a32d60124c13afdfb881e5eec508389b82", - "silhouette.png": "8b4ed63347c4f822a35e8eb66ab76ff1d7f16c377a0495e75bcb171dc6d6ee54", - "value.png": "4080135679b0add47db0a3dc0d695a779cb85772556990ec70bb9b296352a927", - "detail.png": "4911c16326d89110ff23540209898a31827fe3b69936543c15ac0bfbbc751011" + "hero.png": "40450037451d96c91c32b13f78361aed555a4c5809be8ebb84224c7ed0efee93", + "sheet.png": "2d7c16edfdb067b40c78acc91881e9305ec9dc2521a3d038e04bef0bd2234a2d", + "raking.png": "d1760a3c266695055ebff954677f8e53f48cc18b7c01a2e86522f704ee588761", + "raking_detail.png": "0c74e58be9fb3e88219c5068fad66e689b1a0003fca621540e3dbf9d69949c9a", + "lods.png": "08601d760f4eb22cc07959c9af4af743e0fd7ef53d71c79e21b364a59a672e42", + "silhouette.png": "be4d701c94242e7621c8462a2a36f7397e901d9825e0bafbfdf7493c68d2621c", + "value.png": "b7e88d50e9c2152aabdff6008a882465307c945a23d556cf7a83e4d469552ba0", + "detail.png": "ea22466a13fe24ed32d914c01ed54527ed52f1757e0306951705653d7ebf6622" }, "metrics": { "silhouette": [ @@ -75,7 +74,7 @@ }, { "view": "side", - "iou": 0.871 + "iou": 0.949 }, { "view": "3/4", @@ -86,22 +85,22 @@ "peakPart": "turret_body", "peak": [ 403, - 55 + 54 ], "focalShare": 0, "focalArea": 0, "top": [ { "part": "turret_body", - "share": 0.598 + "share": 0.587 }, { "part": "fittings", - "share": 0.202 + "share": 0.199 }, { "part": "hull", - "share": 0.118 + "share": 0.127 } ], "declared": [] diff --git a/recipes/tiger_1/out/tiger_1/critique.json b/recipes/tiger_1/out/tiger_1/critique.json index 848d7bc9..b5911506 100644 --- a/recipes/tiger_1/out/tiger_1/critique.json +++ b/recipes/tiger_1/out/tiger_1/critique.json @@ -1,86 +1,81 @@ { "asset": "tiger_1", "reviewer": "agent", - "overall": 7.1, + "overall": 6.7, "scores": { "silhouette": { "score": 8, - "reason": "Reads as a mid-production Tiger I from every view: slab hull with panniers, interleaved 8-station running gear, horseshoe turret with low cupola, broad 1.9 m mantlet plate with curled ends and gun collar, long 88 with muzzle brake. Rear now has the two tall vertical exhausts in armoured covers, side plates proud of the rear plate, hinged rear mudguards; top view shows the engine deck (centre hatch, mesh radiator grilles, four fan grilles).", + "reason": "Mid-production Tiger I: slab hull with panniers, interleaved running gear, horseshoe turret with low cupola, mantlet, long 88 with muzzle brake, rear exhaust covers. Reads from every view.", "fixes": [ - "round the mantlet's top/bottom edges (cast radius) instead of a straight-extruded plate", - "turret rear bin: rounded rear corners" + "round the mantlet top/bottom edges (cast radius)" ] }, "proportions": { "score": 7, - "reason": "Sheet 8.7 \u00d7 3.6 \u00d7 3.0 m vs dossier 8.45 \u00d7 3.56 \u00d7 3.00: height now right (cupola lowered); length +3% because exhaust covers, pintle and rear mudguards hang behind the 6.316 m hull while the dossier length is hull + gun overhang. All six measures (hull, wheel, track, centres, barrel, station pitch) pass at \u00b12%.", + "reason": "8.7 x 3.6 x 3.0 m against the 8.45 x 3.56 x 3.00 m dossier; measures pass, length +3% from rear fittings.", "fixes": [ - "verify whether the 8.45 m figure includes the exhaust covers; if not, document the 0.25 m difference rather than shortening the gun", - "check the turret overhang beside the Bundesarchiv side photo" + "document whether the 8.45 m figure includes the exhaust covers, or move the covers" ] }, "materials": { - "score": 7, - "reason": "Paint is now steel_painted (Dunkelgelb as vertex colour, chips expose steel), tyres rubber, track rusted manganese steel, tool heads cast iron, handles oak, exhausts rusted iron; UVs box-projected per triangle at true scale. Camo is soft-edged overspray. Weak: chip decals read as uniform dark dots everywhere instead of on edges; Zimmerit ridges have the same paint response as the plate.", + "score": 6, + "reason": "Dunkelgelb with red-brown and green blotches, Zimmerit reads as fine corrugation, rubber tyres; the camo blotches look printed over the Zimmerit rather than sprayed, chips are uniform dark dots, and the wheels read as bronze plastic.", "fixes": [ - "a project material for Zimmerit (rough, matte paste) instead of steel_painted", - "fewer, edge-biased chips: a weathered() steel_painted variant with lower chip density for large plates" + "a matte Zimmerit project material", + "fewer, edge-biased chips", + "darker, rougher steel wheel discs" ] }, "detail": { "score": 7, - "reason": "Zimmerit as 16 mm relief ridges with chipped patches, tow cables with eyes, tools, spare links, cupola periscopes, mantlet bolts and sight apertures, hatch hinges/handles/periscope hoods, engine deck mesh and fan grilles, exhaust shrouds with heat slits and clamp bands, jack, crank cover, lights, pintle. Missing: weld beads at plate interlocks, track-guard brackets, sprocket teeth hidden in SIDE view.", + "reason": "Zimmerit relief, tow cables, tools, spare links on the turret, cupola periscopes, mantlet bolts, engine-deck grilles, exhaust shrouds, lights, number and cross.", "fixes": [ - "weld beads along the interlocked hull corners", - "fender support brackets every ~1 m", - "spare-link hangers and tool clamps" + "weld beads along interlocked hull corners", + "fender support brackets" ] }, "wear": { - "score": 7, - "reason": "Ragged-edged mud on tracks, wheels, lower hull and fenders (heavier at the rear), soot plumes from the exhaust mouths onto the deck, heat-stained muzzle brake, chipped Zimmerit, dust lift on upward plates, chipped paint from the strip. Still too uniform on the upper hull; no polished climbing steps or scratches around hatches.", + "score": 5, + "reason": "Raking.png shows little mud on the running gear and none on the lower hull; the dot chips and blotches carry the read.", "fixes": [ - "polished bare metal on hatch rims and climbing steps (aluminium-strip patches)", - "rust streaks under tow-cable clamps and exhausts", - "stronger dust on the deck and turret roof" + "grime({ ground }) mud on tracks, wheels and lower hull, heavier at the rear", + "exhaust soot on the deck" ] }, "lod": { "score": 7, - "reason": "LOD2 (3.4k) keeps the horseshoe turret, interleaved wheels as discs inside an open belt, crosses, '332' and the camo colour blocks; colours match LOD0. Zimmerit and deck furniture drop out so the hull reads smoother; LOD1\u2192LOD2 still drops the mantlet bolts and exhaust covers.", + "reason": "LOD2 (3.5k) keeps the turret, interleaved wheels as discs, crosses, number and camo blocks; Zimmerit and deck furniture drop.", "fixes": [ - "keep the two exhaust covers as boxes at LOD2", - "bake Zimmerit banding into vertex colour at LOD1/LOD2" + "keep the two exhaust covers as boxes at LOD2" ] }, "technical": { "score": 7, - "reason": "QA PASS incl. mass (62.6 t vs 57 t \u00b110%: hull/turret area-weighted 45 mm walls from the armour table, Zimmerit at paste density, dished wheels as 15 mm walls). 72 360 / 18 096 / 3 410 triangles, one material. LOD0 above the 60k a hero tank usually gets; mass sits near the top of its tolerance. Now 73 712 / 18 096 / 3 410. Connections gate clean (was 46 findings): interleaved inner road-wheel discs now ride on a hub/axle stub per station, Zimmerit ridges sink 4 mm into the plate, nose tenons stand on the nose plate, mantlet bolts follow the curved plate ends, AA ring carried on six brackets off the cupola lip.", + "reason": "QA clean (pack): 68.9k / 17.4k / 3.5k tris, one material; LOD0 above the ~60k a hero tank usually gets.", "fixes": [ - "drop LOD0 below 60k: instance the track link as 2 boxes, merge Zimmerit bands", - "close the open radiator/fan shells so mass is not counted as convex hulls (1 800 open shells)" + "drop LOD0 below 60k: instance the track link as 2 boxes, merge Zimmerit bands" ] } }, "renders": { - "hero.png": "a940675c9d41646ceaaa399c0378e04f8110df07bef341e63e9c8f1a3d33847f", - "sheet.png": "ebe6e2cf891b862baa0164574cc47c00b18ab968783058517da7f1fbff0a8219", - "raking.png": "b89772e154fd71f901379d2214ef9cd33f07a07c526b0a878909dda0ea15078f", - "raking_detail.png": "bec67ad38e368c6cccb3799d799fdfe71f012dd431a45a4275e81b427fbc0981", - "lods.png": "1c09882aa7b4d592d96cf30411ab107f16f506b25dfd63f4d3e4b66d1ae0dc8d", - "silhouette.png": "ab969a231e59d567df478cabe2a2bc8dc72891e8efd527823a178155b52d5ad4", - "value.png": "afc0eb4087f5fac1938868555f1f04e6ef62c5fd440e6c088118e2ab25ffb859", - "detail.png": "c5e32d85d27644107154097d105faabba4261dd8ff9abfb66916e8fc32e385ce" + "hero.png": "ac18d9bf90623cd73bb6decafd7e488485b5553537127de93a184866ce86738b", + "sheet.png": "1498c8a40f3a497de31928a2f75d2ebfbeedbaf20e4f6c0f0b1f8048fe74af78", + "raking.png": "a25b652749ea53621b7083495dfc9496f4e703533f21a0c3afddfa09aaf05b86", + "raking_detail.png": "e47b41d9a43c283ce231f7b849d2bf530e67c66ce75d43c60e6cb924fde9f548", + "lods.png": "b6eb3571674c38e399c86f5dd7e9d730e1962196293a75cdcacf25093b9cea53", + "silhouette.png": "0fe13b3df9a3012fd2bca4305d67f77fcec5b8ba5b683812d25410808f7cd32e", + "value.png": "d39ed78c7b1884ae69b2f47d958f26fe3b4f9a8617da5d7a5ce6128915ceb2ce", + "detail.png": "0be24bb2c978eda7c2519f83b2ecf009f4dcf918bb3a738d51e6bab1f6017ae1" }, "metrics": { "silhouette": [ { "view": "front", - "iou": 0.962 + "iou": 0.966 }, { "view": "side", - "iou": 0.901 + "iou": 0.926 }, { "view": "3/4", @@ -88,38 +83,38 @@ } ], "focal": { - "peakPart": "turret_zimmerit", + "peakPart": "hull", "peak": [ - 208, - 122 + 390, + 192 ], "focalShare": 0, "focalArea": 0, "top": [ { "part": "turret_body", - "share": 0.281 + "share": 0.193 }, { "part": "hull", - "share": 0.159 + "share": 0.19 }, { "part": "hull_detail", - "share": 0.152 + "share": 0.149 } ], "declared": [] }, "rest": { - "share": 0.312, - "spread": 0.73, + "share": 0.461, + "spread": 0.748, "min": 0.25 }, "shape": { - "acuteShare": 0.172, - "straightShare": 0.644, - "taper": 0.178, + "acuteShare": 0.178, + "straightShare": 0.651, + "taper": 0.179, "shape": "stable" } } diff --git a/recipes/tune_demo/out/tune_demo/critique.json b/recipes/tune_demo/out/tune_demo/critique.json new file mode 100644 index 00000000..ef1a923d --- /dev/null +++ b/recipes/tune_demo/out/tune_demo/critique.json @@ -0,0 +1,115 @@ +{ + "asset": "tune_demo", + "reviewer": "agent", + "overall": 6, + "scores": { + "silhouette": { + "score": 8, + "reason": "Slim two-handled amphora with a stem foot, high shoulders, tall neck and flared lip; the outline matches the tuned drawing from the front.", + "fixes": [ + "slight lip thickening (rolled rim) so the top edge reads in silhouette" + ] + }, + "proportions": { + "score": 8, + "reason": "0.3 x 0.3 x 0.6 m, tuned to target_front.svg: belly, neck and handle reach read as a real storage amphora.", + "fixes": [ + "fit the side silhouette as well (handles are thin from the side)" + ] + }, + "materials": { + "score": 5, + "reason": "One flat terracotta colour with a soft sheen; no clay texture, throw rings or colour variation.", + "fixes": [ + "terracotta strip with throw-ring relief and a slightly darker, rougher foot" + ] + }, + "detail": { + "score": 4, + "reason": "Body, foot, lip and two handles; nothing else (no stamp, seam, handle attachment pads).", + "fixes": [ + "handle attachment pads where the handles meet the neck and shoulder", + "a potter's stamp on the shoulder" + ] + }, + "wear": { + "score": 3, + "reason": "Pristine: no chips on the lip, no dirt in the foot, no stains.", + "fixes": [ + "chipped lip and worn foot ring", + "dark residue stain below the lip" + ] + }, + "lod": { + "score": 6, + "reason": "FAR LOD2 (272 tris) keeps the outline but the body becomes visibly 12-sided and the handles become angular polylines.", + "fixes": [ + "16 sides at LOD2 within the 272-triangle budget by fewer profile stations" + ] + }, + "technical": { + "score": 8, + "reason": "QA clean (pack): 4.1k / 1.2k / 272 tris; tune.json params read by every build.", + "fixes": [ + "fewer LOD0 profile stations on the straight neck" + ] + } + }, + "renders": { + "hero.png": "e60676ac44025bd1c0aa4428d7a499b7539e934f9fc5d5743785904f16b3d149", + "sheet.png": "5c57651ca55b3193d26154d8aee153c149d2a525e132c51e4769cdf5ffcf9c25", + "raking.png": "94c59969cd9d316fd35dd3c1dbb7fa032815a8f25b4046bbf72310ece667d47a", + "raking_detail.png": "7bbf0182d78ff9c63fbf3f07e3aa3a898ebce17def1c3df2f626ddf5b6ff39d8", + "lods.png": "5fc5342d99248392c927e4498143d7578a9f53c17eb1eec10dc64faa4ef72454", + "silhouette.png": "9777359d8d7b2315f3e1b0a61dcabae8fc2a74434bd5e17d94f0545310df7de7", + "value.png": "a79553ffbb9861e913df08c1e1181a066a3e548be778777b40fed25282148a21", + "detail.png": "3249dfed1ca0bb8b49b4a41169615ba1507bc2a509a2be1313a33ab183c8fa50" + }, + "metrics": { + "silhouette": [ + { + "view": "front", + "iou": 0.968 + }, + { + "view": "side", + "iou": 0.968 + }, + { + "view": "3/4", + "iou": 0.956 + } + ], + "focal": { + "peakPart": "body", + "peak": [ + 352, + 151 + ], + "focalShare": 0, + "focalArea": 0, + "top": [ + { + "part": "body", + "share": 0.769 + }, + { + "part": "handle_r", + "share": 0.231 + } + ], + "declared": [] + }, + "rest": { + "share": 0.848, + "spread": 0.533, + "min": 0.25 + }, + "shape": { + "acuteShare": 0.599, + "straightShare": 0.496, + "taper": -0.27, + "shape": "aggressive" + } + } +} diff --git a/recipes/type_specimen/out/type_specimen/critique.json b/recipes/type_specimen/out/type_specimen/critique.json new file mode 100644 index 00000000..4eaf67fb --- /dev/null +++ b/recipes/type_specimen/out/type_specimen/critique.json @@ -0,0 +1,112 @@ +{ + "asset": "type_specimen", + "reviewer": "agent", + "overall": 6.4, + "scores": { + "silhouette": { + "score": 7, + "reason": "Free-standing notice board on two posts with a capping rail: reads as a notice board; the outline is a generic rectangle.", + "fixes": [ + "a small gabled hood or bracketed cap over the board" + ] + }, + "proportions": { + "score": 8, + "reason": "0.2 x 3.3 x 2.8 m; 3.2 x 1.9 m face at ~0.8 m off the ground, cap heights per README; lettering scale reads right.", + "fixes": [ + "posts slightly heavier (0.12 m) for a 3.2 m board" + ] + }, + "materials": { + "score": 6, + "reason": "Dark green painted board with faint streaks, cream painted letters and brass raised letters separate well; the posts and frame are flat dark brown with no grain.", + "fixes": [ + "wood grain on posts and frame (wood_painted)", + "brass letters with a polished highlight on the faces" + ] + }, + "detail": { + "score": 6, + "reason": "Six faces twice (flat and raised), title, brass divider, labels; no fixings, bolts or a back panel structure.", + "fixes": [ + "bolts at the post-to-board joint", + "cross battens on the board back" + ] + }, + "wear": { + "score": 5, + "reason": "Faint vertical streaks on the green paint; no fading, no chipped letters, no damp at the post feet.", + "fixes": [ + "sun-faded top band and chipped edges on the frame", + "damp/mud at the post feet" + ] + }, + "lod": { + "score": 6, + "reason": "FAR LOD2 drops the whole left (flat) column and its labels, so half the board content vanishes at distance; title and the raised column stay.", + "fixes": [ + "keep the flat column at LOD2 at low segment count (text() segments) instead of dropping it" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean (pack): 38.5k / 27.8k / 4.1k tris; LOD1 is barely lighter than LOD0, so lettering triangles are not stepped down.", + "fixes": [ + "lower text() segments at LOD1 (half of LOD0)" + ] + } + }, + "renders": { + "hero.png": "5b7563f339529937d0863f0c11a016a716f9bbb00cfe2ea2fc55d763aab9b280", + "sheet.png": "b5408ba2c91374b1f4710330779ca26645ea41b71225e4e3ef28313749fa45cf", + "raking.png": "e68525b7e2fd24c5b07e98112b5d472b4c551a6274d51e108e2c9fcf7978c700", + "raking_detail.png": "d5071e8760f99fa2b95f61ae14a31e83f799d0314ed73a9647320129d6b0f26d", + "lods.png": "93d05a297a973fcb723351609937c3a661252e078c3a1b0bce9193e6a3d7da3b", + "silhouette.png": "493236aa38cdebe1f66ebb03f9eea5ee520a58b1aebf776f9ba403bc0e220923", + "value.png": "6d4e05dde58d23c613af963e26189d3930a42ad93779208130047f0706bf6ea4", + "detail.png": "0cd2abeae169a84436bc47662e58a22b7fa3cf5f09a8c63fd289878b2ec9923c" + }, + "metrics": { + "silhouette": [ + { + "view": "front", + "iou": 1 + }, + { + "view": "side", + "iou": 1 + }, + { + "view": "3/4", + "iou": 1 + } + ], + "focal": { + "peakPart": "board", + "peak": [ + 296, + 180 + ], + "focalShare": 0, + "focalArea": 0, + "top": [ + { + "part": "board", + "share": 1 + } + ], + "declared": [] + }, + "rest": { + "share": 0.807, + "spread": 0.629, + "min": 0.25 + }, + "shape": { + "acuteShare": 0.042, + "straightShare": 0.947, + "taper": -0.757, + "shape": "stable" + } + } +} diff --git a/recipes/water_tower/out/water_tower/critique.json b/recipes/water_tower/out/water_tower/critique.json index a5442642..e82672fd 100644 --- a/recipes/water_tower/out/water_tower/critique.json +++ b/recipes/water_tower/out/water_tower/critique.json @@ -1,39 +1,42 @@ { "asset": "water_tower", "reviewer": "agent", - "overall": 7.0, + "overall": 6.4, "scores": { "silhouette": { "score": 8, - "reason": "Classic railroad water tank: octagonal hipped roof with finial, round stave tank, eight-post trestle with X bracing on concrete footings, swinging spout with chain, outlet core. Reads from every view.", + "reason": "Railroad water tank: hipped roof with finial, round stave tank, trestle with X bracing on footings, swinging spout. Reads from every view.", "fixes": [ - "add the float-gauge board and its pulley on the tank side" + "float-gauge board and pulley on the tank side" ] }, "proportions": { - "score": 8, - "reason": "Overall box and the NRHP-sourced measures pass; tank-to-trestle ratio matches the dossier photos.", - "fixes": [] + "score": 7, + "reason": "11.3 x 7.5 x 10.6 m including the spout; tank-to-trestle ratio matches the dossier. The roof is six-sided in the top view and the centre outlet core is very thick versus the posts.", + "fixes": [ + "an eight-sided roof over the round tank, and slim the centre outlet core ~30%" + ] }, "materials": { - "score": 6, - "reason": "Staves now pine grain darkened to oiled cypress, trestle weathered fir, rusted iron hoops and spout fittings, concrete footings, slate shingles. But the tank's tiled UVs run the planks around the circumference, so the staves read as a tile grid rather than vertical boards, and the hoops lose contrast.", + "score": 5, + "reason": "The tank reads as horizontal boards (a tile grid running around the circumference) instead of vertical staves, and the iron hoops are lost against the wood; trestle, footings and roof read correctly.", "fixes": [ - "map the tank with U up the staves (vertical grain) and one board per stave", - "hoops as galvanised or black iron_cast so they read against the wood" + "map the tank with U up the staves (vertical grain, one board per stave)", + "hoops as black iron_cast so they read against the wood" ] }, "detail": { - "score": 7, - "reason": "Staves, hoops with lugs, joists, cap ring, ladder, spout with counterweight and chain, finial. Missing tie-rod turnbuckles on the bracing and the float gauge.", + "score": 6, + "reason": "Staves, hoops, joists, cap ring, ladder, spout with counterweight and chain, finial; missing tie-rod turnbuckles and the float gauge, and the trestle joints have no bolts.", "fixes": [ "turnbuckles on the X rods", + "bolts at trestle joints", "float-gauge board" ] }, "wear": { - "score": 6, - "reason": "Splash-back mud on the footings and lower posts, dust on top faces, weathered (checked, silvered) trestle timber. No water stains under the tank or around the spout, no moss on the footings.", + "score": 5, + "reason": "Silvered trestle timber and some splash on the footings; no water stains under the tank or spout, no moss.", "fixes": [ "dark wet streaks below the tank floor and the spout", "moss on the shaded side of the footings" @@ -41,77 +44,22 @@ }, "lod": { "score": 6, - "reason": "LOD2 (496) keeps tank, roof, posts and spout, but drops the X bracing and its tank turns lighter and stripier than LOD0 (different U mapping on the 8-sided lathe).", + "reason": "FAR LOD2 keeps tank, roof, posts and spout but drops the X bracing, and the tank turns lighter and flat compared with LOD0.", "fixes": [ - "match LOD2 tank UV span to LOD0", - "keep two X braces as flat quads at LOD2" + "match the LOD2 tank colour/UV span to LOD0", + "keep two X braces per side as flat quads at LOD2" ] }, "technical": { "score": 8, - "reason": "QA PASS, 6 616 / 2 816 / 496, two draws (glass-free), lean for a scene prop. Now 7 488 / 3 096 / 496. Connections gate clean (was 14 findings): hoops are closed bands with the tank's own facets and phase, inner face 1 mm into the staves (were round rings 2 cm off the flat staves), lugs bedded on their hoop; the joist grid now ends over the flat cap beams instead of two outer joists hanging past them.", - "fixes": [] + "reason": "QA PASS (rebuilt today): 7.48k / 3.08k / 488 tris, 2 / 2 / 2 draws: lean for a scene prop.", + "fixes": [ + "spend LOD0 triangles on stave edges instead of the roof facets" + ] } }, "renders": { - "hero.png": "b8af0bc6bad7f0a85d07e1add753b2d36d86f51f78cde4056f31353a867e0e63", - "sheet.png": "63f3d002601d749e701282a4cd038600050c4c216555b2dd3b0d182b92d0e189", - "raking.png": "5ad0087b27a0a2389d67af5db1e19518332e98864b8ac030547a7b78d5ece6b5", - "raking_detail.png": "2e6f20451ea65c8f88788deaf547e299251647f8f79814d3700d5ebaacb58381", - "lods.png": "15af2345843322b361568e732840ab5500f66e591304a9e33a27b6dd771658ef", - "silhouette.png": "73260f2f6859efe9c8749f34e57c8e9825321c26f2a08076536a92164e2f8781", - "value.png": "5ab54800f70eee63d1326d3fc1a989a0b4cd065bbc3caca907d6e348e676aac7", - "detail.png": "aa4ef1badd6d4cfc59820735c78ef467877b79d660230b2c85a5426dc709f817" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.909 - }, - { - "view": "side", - "iou": 0.904 - }, - { - "view": "3/4", - "iou": 0.949 - } - ], - "focal": { - "peakPart": "trestle", - "peak": [ - 410, - 370 - ], - "focalShare": 0, - "focalArea": 0, - "top": [ - { - "part": "trestle", - "share": 0.659 - }, - { - "part": "core", - "share": 0.18 - }, - { - "part": "roof", - "share": 0.16 - } - ], - "declared": [] - }, - "rest": { - "share": 0.644, - "spread": 0.66, - "min": 0.25 - }, - "shape": { - "acuteShare": 0.161, - "straightShare": 0.705, - "taper": -0.051, - "shape": "stable" - } + "hero.png": "9ecc9b090959f41af9ec2b10da6f6c0e41070cb6e07ed959cc5eb5d526474e33", + "sheet.png": "deedb7f287a937d2754ab78fba054ebf9fc68ee4ddc02b28955d0f24263bccef" } } diff --git a/recipes/weather_demo/out/weather_demo/critique.json b/recipes/weather_demo/out/weather_demo/critique.json new file mode 100644 index 00000000..6950c1f3 --- /dev/null +++ b/recipes/weather_demo/out/weather_demo/critique.json @@ -0,0 +1,119 @@ +{ + "asset": "weather_demo", + "reviewer": "agent", + "overall": 6.1, + "scores": { + "silhouette": { + "score": 6, + "reason": "Four new/weathered pairs on labelled slabs (door, crate, concrete block, drum): readable, but all simple boxes and cylinders; the weathered copies do not differ in outline (no sag or dents visible).", + "fixes": [ + "stronger crate board sag and a visible dent on the weathered drum so the pairs differ in silhouette" + ] + }, + "proportions": { + "score": 7, + "reason": "Door 0.91 x 2.03 m, drum 0.572 x 0.851 m, crate 0.8 m, block 0.6 x 0.8 x 0.6 m per README.", + "fixes": [ + "label text bigger on the slabs (it is unreadable in hero.png)" + ] + }, + "materials": { + "score": 6, + "reason": "Painted steel door, pine crate, concrete and blue drum read; the drum reads as plastic (uniform gloss), the concrete as clean plaster.", + "fixes": [ + "steel_painted drum with lower gloss and rolled-edge highlights", + "concrete with aggregate and form-tie holes" + ] + }, + "detail": { + "score": 5, + "reason": "Door with frame, hinges and handle, crate boards, block lifting anchors, drum hoops; otherwise plain.", + "fixes": [ + "bung caps on the drum lid", + "nails and corner irons on the crate" + ] + }, + "wear": { + "score": 7, + "reason": "The weathered copies clearly carry rust streaks on the door, rust and dents on the drum, moss and anchor rust bleed on the block, bleached boards and moss on the crate; but the door rust runs as vertical stripes over the whole leaf rather than from the hinge bolts, and the drum rust is blotchy.", + "fixes": [ + "tie the door rust streaks to the hinge bolts and the bottom edge (weather() sources) instead of full-leaf stripes" + ] + }, + "lod": { + "score": 5, + "reason": "lods.png: at LOD2 the weathered door loses its rust and the concrete block loses its moss, so the new/weathered difference, the point of the demo, mostly vanishes at distance.", + "fixes": [ + "bake the weather() colours into LOD2 vertex colour so the weathered copies stay distinct" + ] + }, + "technical": { + "score": 7, + "reason": "QA clean (pack): 95.4k / 21.8k / 3.8k tris for four simple props: weather() tessellation spends most of LOD0.", + "fixes": [ + "coarser weather() tessellation on flat faces at LOD0" + ] + } + }, + "renders": { + "hero.png": "ab612970c5fb24d5ea28fd99d6704926ef981b5724df364e090a0b450e1aa1af", + "sheet.png": "64e4474175e56f9ce155765e8a59beac00157c28f525ee082afdb89f390268b5", + "raking.png": "aaa75be6d27e0489fbe11390e09f149af9ebb8c96d7b9641acb907af267b7b03", + "raking_detail.png": "cda41618415217ef94cf727aef261731cf33e385f9079b3ff2737fa33fed5d7b", + "lods.png": "4eb077217897dcdd1cde34f857d5602fe00ed4a046f07b4034e5596a13fdb949", + "silhouette.png": "b7bbb9640ecc1eebe49e5d997740ef32324189aa2c03ae5643d2fb30226f77a5", + "value.png": "721503176686760349f81e5aa914875adee46d40ca0c52f92034516071eb32b7", + "detail.png": "2b69109c08b8386239eaab6c5af3028015e1f5a1de856f4a67e94780c1d54609" + }, + "metrics": { + "silhouette": [ + { + "view": "front", + "iou": 1 + }, + { + "view": "side", + "iou": 1 + }, + { + "view": "3/4", + "iou": 1 + } + ], + "focal": { + "peakPart": "door_slab", + "peak": [ + 334, + 254 + ], + "focalShare": 0, + "focalArea": 0, + "top": [ + { + "part": "door_slab", + "share": 0.223 + }, + { + "part": "block_old_block", + "share": 0.136 + }, + { + "part": "block_slab", + "share": 0.122 + } + ], + "declared": [] + }, + "rest": { + "share": 0.8, + "spread": 0.574, + "min": 0.25 + }, + "shape": { + "acuteShare": 0.164, + "straightShare": 0.896, + "taper": 0.617, + "shape": "stable" + } + } +} diff --git a/recipes/wolf/out/wolf/critique.json b/recipes/wolf/out/wolf/critique.json index 1173258b..7c3059ee 100644 --- a/recipes/wolf/out/wolf/critique.json +++ b/recipes/wolf/out/wolf/critique.json @@ -1,135 +1,67 @@ { "asset": "wolf", "reviewer": "agent", - "overall": 6, + "overall": 5.9, "scores": { "silhouette": { "score": 7, - "reason": "Side, front and 3/4 views name a grey wolf: deep chest over long legs with the elbow at the chest floor, tucked waist, a visible neck carrying the head forward of and above the shoulders with a throat line from jaw to brisket, broad-based ears, bushy brush tail hanging to the hocks (LOD2 IoU 0.947/0.916/0.921). Against the Commons standing-in-snow side view the mane/ruff barely changes the outline and the winter coat does not break the outline anywhere (no guard-hair tufts at the elbows, throat, thighs). Shape metric reads 'stable' vs declared 'friendly' (MISS) — the trunk box dominates.", + "reason": "Long legs, tucked waist, neck carrying the head forward, upright ears, bushy tail to the hocks and a fur fringe along the back and chest: reads as a wild canid from every view. The narrow, long muzzle, big ears and slim chest read closer to a coyote than a grey wolf.", "fixes": [ - "fur cards or shells along the mane, throat, elbows and breeches so the outline breaks like a winter coat" + "broaden the muzzle and skull and deepen the chest so the head and front read as a grey wolf, not a coyote" ] }, "proportions": { - "score": 7, - "reason": "Sourced numbers hold: head+body 1.15 m (+-2% test), tail 0.42 m, withers 0.80-0.86 m test, mass 30-60 kg test and QA mass +-25%, skull width 0.55 x length, box 1.30 x 0.277 x 0.894 m vs real 1.31 x 0.27 x 0.90. By eye the chest still reads slightly barrel-like in the 3/4 hero and the hind cannon is long; the neck (0.28 m at 40°) is an estimate, not sourced.", + "score": 6, + "reason": "1.3 x 0.3 x 0.9 m fits the sourced head+body and withers; by eye the head is small and the snout long and thin for a wolf, the ears large, and the hind cannons long in the side view.", "fixes": [ - "source the cervical length and chest width from a wolf skeleton reference and pin them in the README table", - "narrow the brisket keel ~10% so the front view reads more wedge than barrel" + "skull width ~0.55 of skull length with a shorter, deeper muzzle", + "ears ~15% smaller", + "shorten the hind cannon toward the dossier skeleton" ] }, "materials": { - "score": 5, - "reason": "Vertex colour on the fur_neutral strip: saddle darker than the buff flanks with no hard edge at the shoulder, cream belly and throat, black tail tip, grey eye mask and pale brows, foreline on the forearms, grey-buff furred toes over dark pads all read. But it is velvet, not fur: the agouti grizzle is soft ~1 cm streaks, no hair clumps, sheen or strand breakup; claws and nose read as plastic; toes go slightly lilac under the cool key.", + "score": 6, + "reason": "Agouti saddle, buff flanks, cream chest and belly, dark tail tip and glossy nose read; strand cards along the back and chest break the outline like fur, but the body between reads as smooth painted velvet, and the grey chest goes lilac under the cool key.", "fixes": [ - "fur shells or cards with a strand alpha (planned in M4)", - "a dedicated wolf coat strip with directional strand normals" + "fur shells or more strand cards over the flanks and shoulders, not only the fringe", + "warm the chest cream so it does not go lilac" ] }, "detail": { - "score": 5, - "reason": "LOD0 has the anatomy landmarks (triceps, forearm, elbow point, carpal pad, quads/hamstrings, calf and Achilles, stifle, hock, knuckled toes with claws and dark pads, stop, flews, cheek fur, lidded eyes, fangs); the fold across the withers is gone (loin runs under the ribcage, tapered mane). At third-person distance the coat surface is still smooth and the mane a soft swell.", + "score": 6, + "reason": "Eyes with lids, nose, lip line, knuckled toes with claws and pads, ear interiors, chest ruff; the coat surface has no hair-flow relief at hero distance, and the face has no cheek ruff or brow fur.", "fixes": [ - "guard-hair relief strokes following the hair flow", - "separate the mane into a few overlapping locks" + "guard-hair relief following the hair flow on the flanks", + "cheek ruff and brow tufts on the head" ] }, "wear": { "score": 4, - "reason": "No story: clean coat, no matted or wet fur, no scars, no mud on the paws or belly, symmetric apart from the seeded grizzle.", + "reason": "No story: clean coat, no matted or wet fur, no mud on the paws or belly, no scars.", "fixes": [ - "grime({ ground }) on paws and lower legs", - "a few darker matted streaks on the flank and a nick in one ear" + "grime({ ground }) mud on paws and lower legs", + "darker matted streaks on the flank and a nick in one ear" ] }, "lod": { - "score": 7, - "reason": "lods.png: LOD2 keeps the outline (IoU 0.93-0.95) and the big colour blocks; legs become 6-sided tubes and the paws rounded spindles (no longer boxes, darkened to match). The grizzle and eye detail drop, and the ears thin to dark-edged blades at LOD2.", + "score": 5, + "reason": "lods.png: LOD2 (2.3k) keeps the pose and outline, but the fur fringe and tail volume go, legs become pale faceted tubes with boxy paws, and the saddle colour washes out to one beige.", "fixes": [ - "8-sided leg tubes at LOD2 within the 4000 budget", + "8-sided leg tubes and rounded paws at LOD2 within budget", "paint the agouti saddle as a darker block at LOD2 so the value pattern holds" ] }, "technical": { - "score": 8, - "reason": "QA PASS at every LOD (connections incl. claws rooted in the toes, mass, balance, proportions, normals); triangles 139k / 29k / 3.0k inside the 150k / 35k / 4k budget; one draw call; deterministic. Most LOD0 triangles go to smooth fur areas that do not need them.", + "score": 7, + "reason": "QA clean (pack): 141.7k / 25.9k / 2.3k tris, inside the 150k / 35k / 4k budget; most LOD0 triangles go to smooth flank areas that do not need them. The pack hero re-render drifted from the committed hero.png bytes (visually identical).", "fixes": [ - "adaptive cell size: finer on the head and paws, coarser on the flanks" + "adaptive cell size: finer on the head and paws, coarser on the flanks", + "re-run bun run shots wolf so the committed hero matches a fresh render byte for byte" ] } }, "renders": { - "hero.png": "bbd210744390d7681d6d96bfde915e9669a1deef5a777f5a4ef76bd636311b55", - "sheet.png": "c8edcb9dd7813d8a2e5884c3dbfbae5f7d4ca59a1bc113de232412ae2e474336", - "raking.png": "1702ddcf2f9ddf5d03bdb5720a5d43759a21e184258c657f4e16bc449fd244c8", - "raking_detail.png": "3ead09072d112ed02e3744d961aa005cef1db0342e000aff45cb42b029f07b0d", - "lods.png": "b72d1e3347eaad33935e94515ea11870d8e35a04615d9db376278869b84c2850", - "silhouette.png": "ba8a3d430ca1a9c7b83b58226b258462c00ce72e2c3ce0b6dafeb098f646fe8d", - "value.png": "44c7d6a775a86560c1835221061a2d1a377fb4288977ec813016fb377004553d", - "detail.png": "6880583723fc70669e711867d8bb52bc50f05cd1058900925c59b7998a75ca57" - }, - "metrics": { - "silhouette": [ - { - "view": "front", - "iou": 0.947 - }, - { - "view": "side", - "iou": 0.916 - }, - { - "view": "3/4", - "iou": 0.921 - } - ], - "focal": { - "peakPart": "body", - "peak": [ - 362, - 291 - ], - "focalShare": 0.392, - "focalArea": 0.135, - "top": [ - { - "part": "body", - "share": 0.568 - }, - { - "part": "head", - "share": 0.392 - }, - { - "part": "eye_r", - "share": 0.019 - } - ], - "declared": [ - "head" - ] - }, - "rest": { - "share": 0.55, - "spread": 0.549, - "min": 0.25 - }, - "grounded": { - "rise": 0.059, - "bands": [ - 0.359, - 0.377, - 0.441, - 0.386, - 0.413 - ] - }, - "shape": { - "acuteShare": 0.357, - "straightShare": 0.555, - "taper": -0.55, - "shape": "stable", - "declared": "friendly" - } + "hero.png": "082d00756a01060b95bb56375596149a91aafc8e85efc410e3c706dfa2368385", + "sheet.png": "a6431e8ff36861d8087a070d41f5f9b08a277c3815f0b2554531f16e7ec4f9bf" } } diff --git a/recipes/wolf_hybrid/README.md b/recipes/wolf_hybrid/README.md index ec266c7c..803c2724 100644 --- a/recipes/wolf_hybrid/README.md +++ b/recipes/wolf_hybrid/README.md @@ -35,6 +35,12 @@ The check reads LOD 0, like `real.measures`: a decimated far LOD loses a millime Five cached requests, each naming the previous output by sha256 (kinds `modelChain`). A build never calls a model or Blender; `t3d neural run` does. +The chain is `required: true`. The code wolf this spec would give without the model is not this +asset (over 100 000 triangles at LOD 0 against a 30 000 budget, off the sourced box), so `t3d build` +with an output missing stops with one `T3D_NEURAL_MISSING` and builds nothing. The npm package does +not ship `neural/`: clone the repository to build this recipe, or copy the folder into your project +and run `t3d neural run` on the copy (GPU and Blender, below). + | # | Task | Tool (licence) | Output | In git | Time (RTX 3090) | |---|---|---|---|---|---| | 1 | `image` | Z-Image-Turbo (Apache-2.0), seed 9, 1024² | concept PNG, 0.96 MB | yes | 1 min 58 s (41 s of it loading the model) | @@ -129,7 +135,7 @@ leg (GPU). Then lower the ratchet to 0 and drop both tolerances. | Surface | code body moved ≤ 12 mm toward the model, colour by closest point | the model's own mesh, retopologised | | Texture | vertex colour | baked albedo, normal, ORM | | Joints | plan | UniRig, mapped to the plan | -| Fallback | code wolf | code wolf (unchanged) | +| Without the cache | code wolf | `T3D_NEURAL_MISSING` in a build (the chain is required) | ## Re-roll diff --git a/recipes/wolf_hybrid/out/wolf_hybrid/critique.json b/recipes/wolf_hybrid/out/wolf_hybrid/critique.json new file mode 100644 index 00000000..7363f0db --- /dev/null +++ b/recipes/wolf_hybrid/out/wolf_hybrid/critique.json @@ -0,0 +1,63 @@ +{ + "asset": "wolf_hybrid", + "reviewer": "agent", + "overall": 7.0, + "scores": { + "silhouette": { + "score": 8, + "reason": "Reads as a grey wolf from every view: long muzzle, pricked ears, deep chest, tucked belly, long straight legs and a hanging brush tail; the SIDE view in sheet.png gives the standing-wolf outline at once.", + "fixes": [ + "give the neck ruff a broken, tufted outline (a short shell or card pass) so the head does not join the shoulders in one smooth curve" + ] + }, + "proportions": { + "score": 8, + "reason": "1.2 × 0.3 × 0.9 m on the grid sits inside the sourced adult range; leg length, chest depth and head size look right beside photos. The tail hangs to the hocks, at the long end of the range.", + "fixes": [ + "shorten the tail a few centimetres in a Blender step so it ends above the hock, as on most reference photos" + ] + }, + "materials": { + "score": 7, + "reason": "The image-to-3D texture reads as a grizzled grey coat with a pale chest and legs, and the face markings hold. Lighting is baked into the albedo (bright back, saturated orange flank and ear patches) and the fur has one uniform roughness.", + "fixes": [ + "de-light the albedo before import (a bake step that divides out the baked sun and AO)", + "pull the orange patches on the flank, ears and tail toward the coat's tawny grey" + ] + }, + "detail": { + "score": 7, + "reason": "Fur breakup, eyes, nose and toes come from the texture and hold at third-person distance. Geometry is smooth: no fur clumps on the outline, and the flank shows rib-like ridges in SIDE that read as a starved animal rather than a coat.", + "fixes": [ + "smooth the ridges on the flank in a Blender sculpt step", + "a strand or shell pass on the ruff, back and tail so the outline breaks into fur clumps" + ] + }, + "wear": { + "score": 5, + "reason": "A clean animal: the darker back gives a little story, but there is no mud on the paws or lower legs, no wet or matted fur, no scars.", + "fixes": [ + "mud on the paws and lower legs (grime with the ground mask over the textured part)", + "a seeded scar on the muzzle or a nicked ear" + ] + }, + "lod": { + "score": 8, + "reason": "FAR LOD2 keeps the outline, the pale chest and the dark back; at the sheet's far distance it is hard to tell from LOD0.", + "fixes": [ + "lower the LOD2 triangle count: the far view does not need its current budget for a 1.2 m animal" + ] + }, + "technical": { + "score": 6, + "reason": "QA PASS and one draw per LOD, with idle, look and walk clips, but only with a declared deform tolerance: the skin webs between the tail root and the thighs at the hip's extremes (stretch up to ×5.7 against a limit of ×4, defect D253, still open).", + "fixes": [ + "close D253: a per-bone reach or blend at the tail root in the `isolate` Blender step, or a regenerated mesh from a concept with one raised tail, then remove `limitTolerances`" + ] + } + }, + "renders": { + "hero.png": "74112f24b1c1bf6117f3dc336b607f9e3465e088338ef3690e9db710c13772b0", + "sheet.png": "00f3f340f30370bb375acdeb36ceca17dba6ec4a5b3182a3ce266ef4d8a3035f" + } +} diff --git a/recipes/wolf_hybrid/surface.ts b/recipes/wolf_hybrid/surface.ts index 8db621b3..e47c4b52 100644 --- a/recipes/wolf_hybrid/surface.ts +++ b/recipes/wolf_hybrid/surface.ts @@ -9,7 +9,10 @@ // elbows and the spine lie (mapped by topology) // blender FINAL: transform(fit) + steps.ts → the game mesh: three LODs, PNG maps, the skin; // committed, the only thing a build draws -// A miss anywhere returns null: the code wolf (wolf_hybrid.ts). Nothing here calls a model; +// The chain is `required`: the code creature of this spec fails QA (over 100 000 triangles at +// LOD 0, off the sourced box), so `t3d build` with an output missing stops with one +// T3D_NEURAL_MISSING and never ships it. A compile with no cache provided at all (a test, the +// gallery's CPU tier) still gets null: the code wolf (wolf_hybrid.ts). Nothing here calls a model; // `t3d neural run` does, on a machine with a GPU and Blender (README.md). import { @@ -39,7 +42,9 @@ export const SURFACE: Omit = { * behind, and this model's tail was generated lying on the rump and hocks. */ export const DROPPED: readonly string[] = ['gallop']; -/** The imported game mesh with the rig it was built for, or null on any cache miss. */ +/** The imported game mesh with the rig it was built for; null with no cache provided + * and while `t3d neural run` is still filling the cache; an offline (build) cache throws + * `T3D_NEURAL_MISSING` on a miss instead. */ export const wolfModel = (ctx: BuildContext): ChainedModel | null => modelChain(ctx, { concept: CONCEPT, @@ -52,6 +57,7 @@ export const wolfModel = (ctx: BuildContext): ChainedModel | null => maxTriangles: TRIANGLES[0], models: { image: CONCEPT_MODEL, mesh: SURFACE_MODEL }, license: 'MIT', + required: true, author: 'ts23d wolf_hybrid recipe (Z-Image concept, Pixal3D mesh, UniRig joints, Blender steps)', }); diff --git a/recipes/wolf_hybrid/wolf_hybrid.test.ts b/recipes/wolf_hybrid/wolf_hybrid.test.ts index 8475e307..5573dc8c 100644 --- a/recipes/wolf_hybrid/wolf_hybrid.test.ts +++ b/recipes/wolf_hybrid/wolf_hybrid.test.ts @@ -7,7 +7,14 @@ import { afterEach, describe, expect, it } from 'bun:test'; import { join } from 'node:path'; -import { clearNeuralSources, compile, runQa, toGlb } from '@developerz.ai/ts23d'; +import { + clearNeuralSources, + compile, + neuralCache, + provideNeural, + runQa, + toGlb, +} from '@developerz.ai/ts23d'; import { compileLods } from '@developerz.ai/ts23d/testing'; import { provideRecipeNeural } from '../../src/cli/neural-store.ts'; import { TRIANGLES } from './steps.ts'; @@ -39,6 +46,12 @@ describe('wolf_hybrid, CPU tier (no neural cache)', () => { }); describe('wolf_hybrid, model tier (committed cache)', () => { + it('a build with an empty cache (the npm package) stops with T3D_NEURAL_MISSING, no CPU stand-in', () => { + clearNeuralSources(); + provideNeural(neuralCache({ recipe: 'wolf_hybrid', offline: true })); + expect(() => compile(wolf_hybrid, { lod: 2 })).toThrow('T3D_NEURAL_MISSING'); + }); + it( 'a missing cache is not mistaken for the model: the textured body appears only with it', async () => { diff --git a/scripts/lib/recipe-costs.json b/scripts/lib/recipe-costs.json index 5f335e15..b6957fc8 100644 --- a/scripts/lib/recipe-costs.json +++ b/scripts/lib/recipe-costs.json @@ -1,126 +1,127 @@ { - "shard1": 15.8, + "shard1": 25.8, + "overhead": 16.3, "seconds": { - "gate:recipes/amur_tiger": 52.8, - "gate:recipes/asteroid": 9.6, - "gate:recipes/bear_hybrid": 36, - "gate:recipes/bluefin_tuna": 4.2, + "gate:recipes/amur_tiger": 63.3, + "gate:recipes/asteroid": 9.7, + "gate:recipes/bear_hybrid": 36.8, + "gate:recipes/bluefin_tuna": 3.8, "gate:recipes/boulders": 2.3, - "gate:recipes/boxcar": 4.8, - "gate:recipes/brown_bear": 45, - "gate:recipes/cargo_container_scifi": 3.7, - "gate:recipes/city_block": 7.7, - "gate:recipes/common_octopus": 14.6, - "gate:recipes/dragon": 66.7, - "gate:recipes/dragon_hybrid": 20.7, - "gate:recipes/dropship": 4.7, - "gate:recipes/eiffel_tower": 13.2, - "gate:recipes/elf_head": 76.4, - "gate:recipes/face_demo": 173.7, - "gate:recipes/finish_demo": 1.7, - "gate:recipes/fox": 84.7, - "gate:recipes/goblin_head": 58.2, - "gate:recipes/great_white": 4.4, - "gate:recipes/human_head": 147, - "gate:recipes/hydra": 52, - "gate:recipes/import_demo": 10.7, - "gate:recipes/joint_demo": 1.9, - "gate:recipes/king_cobra": 5.3, - "gate:recipes/loco_440": 3.9, - "gate:recipes/material_swatches": 2.4, - "gate:recipes/mountain_range": 13.1, - "gate:recipes/oak": 9.8, - "gate:recipes/orc_head": 80.6, - "gate:recipes/palm": 2.5, - "gate:recipes/parts_catalogue": 2.8, - "gate:recipes/pbr_showcase": 1.9, - "gate:recipes/pine": 17.7, - "gate:recipes/ranger": 16.3, - "gate:recipes/real_valley": 4.5, - "gate:recipes/roadster_v10": 39.5, - "gate:recipes/rts_barracks": 14.4, - "gate:recipes/rts_faction": 10.5, - "gate:recipes/rts_resource": 3.9, - "gate:recipes/rts_soldier": 1.9, - "gate:recipes/rts_tank": 2.8, - "gate:recipes/rts_worker": 1.8, - "gate:recipes/signal_gantry": 4.4, - "gate:recipes/sound_demo": 2.6, - "gate:recipes/station_halt": 2.2, - "gate:recipes/svg_showcase": 3.3, - "gate:recipes/t34": 7.9, - "gate:recipes/tiger_1": 6.7, - "gate:recipes/tune_demo": 2, - "gate:recipes/type_specimen": 3.7, - "gate:recipes/water_tower": 2.4, - "gate:recipes/weather_demo": 4, - "gate:recipes/wolf": 107.8, - "gate:recipes/wolf_hybrid": 25.2, - "recipes/_lib/bust/cloth-check.test.ts": 7.1, - "recipes/_lib/bust/cloth-sim.test.ts": 0.3, - "recipes/_lib/bust/cloth.test.ts": 0, - "recipes/_lib/bust/creases.test.ts": 0, + "gate:recipes/boxcar": 4.2, + "gate:recipes/brown_bear": 37.3, + "gate:recipes/cargo_container_scifi": 4.4, + "gate:recipes/city_block": 7.2, + "gate:recipes/common_octopus": 17.2, + "gate:recipes/dragon": 55.3, + "gate:recipes/dragon_hybrid": 17.2, + "gate:recipes/dropship": 4.8, + "gate:recipes/eiffel_tower": 10.9, + "gate:recipes/elf_head": 71.8, + "gate:recipes/face_demo": 166, + "gate:recipes/finish_demo": 1.5, + "gate:recipes/fox": 86.5, + "gate:recipes/goblin_head": 41.6, + "gate:recipes/great_white": 5.1, + "gate:recipes/human_head": 148.8, + "gate:recipes/hydra": 45.1, + "gate:recipes/import_demo": 11.3, + "gate:recipes/joint_demo": 2.2, + "gate:recipes/king_cobra": 5.9, + "gate:recipes/loco_440": 3.5, + "gate:recipes/material_swatches": 2.8, + "gate:recipes/mountain_range": 13.3, + "gate:recipes/oak": 9.4, + "gate:recipes/orc_head": 69.9, + "gate:recipes/palm": 1.9, + "gate:recipes/parts_catalogue": 3.3, + "gate:recipes/pbr_showcase": 1.8, + "gate:recipes/pine": 18.7, + "gate:recipes/ranger": 14.1, + "gate:recipes/real_valley": 3.4, + "gate:recipes/roadster_v10": 34.3, + "gate:recipes/rts_barracks": 10.3, + "gate:recipes/rts_faction": 12.1, + "gate:recipes/rts_resource": 4.7, + "gate:recipes/rts_soldier": 2, + "gate:recipes/rts_tank": 3.4, + "gate:recipes/rts_worker": 1.3, + "gate:recipes/signal_gantry": 4.9, + "gate:recipes/sound_demo": 2.3, + "gate:recipes/station_halt": 1.9, + "gate:recipes/svg_showcase": 2.9, + "gate:recipes/t34": 8.3, + "gate:recipes/tiger_1": 7.9, + "gate:recipes/tune_demo": 2.2, + "gate:recipes/type_specimen": 3.5, + "gate:recipes/water_tower": 2.6, + "gate:recipes/weather_demo": 3.5, + "gate:recipes/wolf": 95.7, + "gate:recipes/wolf_hybrid": 30.2, + "recipes/_lib/bust/cloth-check.test.ts": 5.1, + "recipes/_lib/bust/cloth-sim.test.ts": 0.2, + "recipes/_lib/bust/cloth.test.ts": 0.1, + "recipes/_lib/bust/creases.test.ts": 0.1, "recipes/_lib/bust/head.test.ts": 0.4, - "recipes/_lib/bust/skin.test.ts": 3, - "recipes/_lib/face/brows.test.ts": 0.4, + "recipes/_lib/bust/skin.test.ts": 2.6, + "recipes/_lib/face/brows.test.ts": 0.5, "recipes/_lib/face/face.test.ts": 0.4, - "recipes/_lib/face/scalp-check.test.ts": 0, - "recipes/_lib/face/skull.test.ts": 1.7, - "recipes/_lib/sets.test.ts": 9, - "shots:recipes/amur_tiger": 33.7, - "shots:recipes/asteroid": 30.6, - "shots:recipes/bear_hybrid": 53.5, - "shots:recipes/bluefin_tuna": 11.6, - "shots:recipes/boulders": 12.4, - "shots:recipes/boxcar": 23.2, - "shots:recipes/brown_bear": 77.3, - "shots:recipes/cargo_container_scifi": 17, - "shots:recipes/city_block": 19.3, - "shots:recipes/common_octopus": 23.3, - "shots:recipes/dragon": 56.8, - "shots:recipes/dragon_hybrid": 23.1, - "shots:recipes/dropship": 16.8, - "shots:recipes/eiffel_tower": 24.2, - "shots:recipes/elf_head": 134.5, - "shots:recipes/face_demo": 232.9, - "shots:recipes/finish_demo": 12.4, - "shots:recipes/fox": 103, - "shots:recipes/goblin_head": 106.5, - "shots:recipes/great_white": 14.5, - "shots:recipes/human_head": 239.2, - "shots:recipes/hydra": 44.3, - "shots:recipes/import_demo": 21.3, - "shots:recipes/joint_demo": 9.4, - "shots:recipes/king_cobra": 14, - "shots:recipes/loco_440": 15.5, - "shots:recipes/material_swatches": 13.6, - "shots:recipes/mountain_range": 38.4, - "shots:recipes/oak": 20.8, - "shots:recipes/orc_head": 98.4, - "shots:recipes/palm": 10.9, - "shots:recipes/parts_catalogue": 13, - "shots:recipes/pbr_showcase": 11.2, - "shots:recipes/pine": 20.9, - "shots:recipes/ranger": 21, - "shots:recipes/real_valley": 16.8, - "shots:recipes/roadster_v10": 78.5, - "shots:recipes/rts_barracks": 68, - "shots:recipes/rts_faction": 19.1, - "shots:recipes/rts_resource": 18.8, - "shots:recipes/rts_soldier": 11.1, - "shots:recipes/rts_tank": 14.1, - "shots:recipes/rts_worker": 11.3, - "shots:recipes/signal_gantry": 17.3, - "shots:recipes/sound_demo": 12, - "shots:recipes/station_halt": 13.4, - "shots:recipes/svg_showcase": 13.9, - "shots:recipes/t34": 28.6, - "shots:recipes/tiger_1": 19.5, - "shots:recipes/tune_demo": 10.5, - "shots:recipes/type_specimen": 12.7, - "shots:recipes/water_tower": 12.1, - "shots:recipes/weather_demo": 14.9, - "shots:recipes/wolf": 95.8, - "shots:recipes/wolf_hybrid": 25.3 + "recipes/_lib/face/scalp-check.test.ts": 0.1, + "recipes/_lib/face/skull.test.ts": 1.9, + "recipes/_lib/sets.test.ts": 7.5, + "shots:recipes/amur_tiger": 36.4, + "shots:recipes/asteroid": 45.5, + "shots:recipes/bear_hybrid": 40.1, + "shots:recipes/bluefin_tuna": 13.5, + "shots:recipes/boulders": 21.1, + "shots:recipes/boxcar": 46.3, + "shots:recipes/brown_bear": 62.9, + "shots:recipes/cargo_container_scifi": 26.9, + "shots:recipes/city_block": 31.2, + "shots:recipes/common_octopus": 26.5, + "shots:recipes/dragon": 45.7, + "shots:recipes/dragon_hybrid": 25.4, + "shots:recipes/dropship": 21.8, + "shots:recipes/eiffel_tower": 27.4, + "shots:recipes/elf_head": 81.5, + "shots:recipes/face_demo": 170.5, + "shots:recipes/finish_demo": 17.5, + "shots:recipes/fox": 48.3, + "shots:recipes/goblin_head": 64.9, + "shots:recipes/great_white": 21.1, + "shots:recipes/human_head": 102.1, + "shots:recipes/hydra": 34.3, + "shots:recipes/import_demo": 26.1, + "shots:recipes/joint_demo": 12, + "shots:recipes/king_cobra": 20.4, + "shots:recipes/loco_440": 30.3, + "shots:recipes/material_swatches": 25.8, + "shots:recipes/mountain_range": 71.4, + "shots:recipes/oak": 26.7, + "shots:recipes/orc_head": 56.4, + "shots:recipes/palm": 12.8, + "shots:recipes/parts_catalogue": 24.4, + "shots:recipes/pbr_showcase": 19.8, + "shots:recipes/pine": 19.5, + "shots:recipes/ranger": 30.3, + "shots:recipes/real_valley": 17.4, + "shots:recipes/roadster_v10": 131.2, + "shots:recipes/rts_barracks": 152.6, + "shots:recipes/rts_faction": 23.6, + "shots:recipes/rts_resource": 26.5, + "shots:recipes/rts_soldier": 14.6, + "shots:recipes/rts_tank": 18.6, + "shots:recipes/rts_worker": 19.4, + "shots:recipes/signal_gantry": 29.6, + "shots:recipes/sound_demo": 21.3, + "shots:recipes/station_halt": 17.7, + "shots:recipes/svg_showcase": 23, + "shots:recipes/t34": 41.5, + "shots:recipes/tiger_1": 25.9, + "shots:recipes/tune_demo": 17.5, + "shots:recipes/type_specimen": 15.2, + "shots:recipes/water_tower": 22, + "shots:recipes/weather_demo": 25.2, + "shots:recipes/wolf": 67.8, + "shots:recipes/wolf_hybrid": 36.1 } } diff --git a/scripts/lib/shard.ts b/scripts/lib/shard.ts index 8cc9a383..db2a0e31 100644 --- a/scripts/lib/shard.ts +++ b/scripts/lib/shard.ts @@ -1,9 +1,9 @@ // CI splits the slow per-recipe work across parallel jobs: T3D_SHARD=k/n (1-based) owns one of n // slices. Unset = everything, so `bun run verify` locally is the whole gate. The slices are -// balanced by measured cost (recipe-costs.json: seconds each folder's gate, each folder's renders -// and each shared recipe test add to a CI shard job; refresh with `bun run scripts/recipe-costs.ts`): -// greedy, longest first, -// into the least-loaded shard, with shard 1 preloaded by the once-only checks it also runs. +// balanced by measured cost (recipe-costs.json: the GitHub seconds each folder's gate, each folder's +// renders and each shared recipe test add to a CI shard job; refresh from a green run with +// `bun run scripts/recipe-costs.ts --from-run `): greedy, longest first, into the least-loaded +// shard, with shard 1 preloaded by the once-only checks it also runs. // Deterministic: the same keys and costs give the same split on every runner. import costsJson from './recipe-costs.json' with { type: 'json' }; @@ -17,13 +17,15 @@ export interface Shard { export interface Costs { /** Seconds shard 1 spends on the steps only it runs (lint, typecheck, build, pack…). */ readonly shard1: number; + /** Seconds every job spends outside its units: checkout, setup, what changed, the stale check. */ + readonly overhead?: number; readonly seconds: Readonly>; } export const COSTS: Costs = costsJson; /** CI's shard count (.github/workflows/ci.yml's gate matrix; scripts/ci-parity.test.ts holds them together). */ -export const CI_SHARDS = 16; +export const CI_SHARDS = 18; /** Parse `k/n`; undefined when unset, a thrown message (with the fix) when malformed. */ export function parseShard(raw: string | undefined): Shard | undefined { diff --git a/scripts/lib/test-timings.json b/scripts/lib/test-timings.json index 061385c4..bac14508 100644 --- a/scripts/lib/test-timings.json +++ b/scripts/lib/test-timings.json @@ -1,479 +1,493 @@ { "version": 1, "files": { - "src/kinds/vegetation.test.ts": 29409, - "src/kinds/creature-family.test.ts": 26178, - "src/cli/bin.test.ts": 18460, - "src/primitives/sculpt-mesh-field.test.ts": 11635, - "src/kinds/model-body.test.ts": 11295, - "src/finish/hero.test.ts": 10671, - "src/testing/readability.test.ts": 10611, - "examples/build-with-bun.test.ts": 8444, - "src/cli/finish-cmd.test.ts": 8217, - "scripts/ci-parity.test.ts": 8153, - "src/qa/connections-seam.test.ts": 7571, - "src/cli/run.test.ts": 7519, - "src/cli/run-json.test.ts": 7017, - "src/cli/critique-cmd.test.ts": 6643, - "src/testing/tiger.example.test.ts": 6613, - "src/cli/concept-cmd.test.ts": 6446, - "src/kinds/model-fit.test.ts": 6319, - "src/kinds/model-chain-two-pass.test.ts": 6072, - "src/kinds/model-rig.test.ts": 5904, - "src/cli/icon-cmd.test.ts": 4540, - "src/kinds/creature-types.test.ts": 4481, - "src/cli/mcp/server.test.ts": 3551, - "scripts/lib/neural-cache.test.ts": 3262, - "src/kinds/creature-fur.test.ts": 3254, - "src/render/water-floor.test.ts": 3225, - "src/kinds/model-repose.test.ts": 2901, - "src/render/pose-strip.test.ts": 2681, - "src/kinds/creature.test.ts": 2651, - "src/cli/mcp/tools.test.ts": 2607, - "src/neural/blender-session-lifecycle.test.ts": 2540, - "src/kinds/head.test.ts": 2512, - "src/kinds/creature-fur-look.test.ts": 2269, - "src/qa/gaps.test.ts": 2092, - "src/qa/critique-calibrate.test.ts": 2070, - "src/neural/meshy-post.test.ts": 2065, - "src/qa/chain-fft.test.ts": 1854, - "src/testing/assert.test.ts": 1655, - "src/kinds/model-skeleton.test.ts": 1560, - "scripts/material-rows.test.ts": 1492, - "src/qa/connections-grown.test.ts": 1420, - "src/cli/setup-cmd.test.ts": 1321, - "src/cli/server-cmd.test.ts": 1274, - "src/neural/tool-reap.test.ts": 1244, - "src/qa/critique-ref.test.ts": 1181, - "src/render/set-shade.test.ts": 1164, - "src/render/photo-match.test.ts": 1159, - "src/qa/ref-gate.test.ts": 1144, - "src/neural/blender-session.test.ts": 1132, - "src/cli/render-cmds.test.ts": 1108, - "src/kinds/swim-clips.test.ts": 1082, - "src/neural/blender-session-lock.test.ts": 1047, - "src/cli/neural-cmd.test.ts": 1045, - "src/kinds/roof.test.ts": 1039, - "src/audio/engine.test.ts": 1036, - "src/finish/finish.test.ts": 1036, - "src/finish/glb.test.ts": 1033, - "src/primitives/sculpt-stroke.test.ts": 1024, - "src/finish/bake.test.ts": 965, - "src/kinds/neck-clips.test.ts": 946, - "scripts/readme-example.test.ts": 894, - "src/primitives/sdf-detail.test.ts": 887, - "src/cli/neural-input.test.ts": 886, - "src/material/live-sheet.test.ts": 868, - "src/kinds/tentacle-clips.test.ts": 838, - "src/material/paint/library-relief.test.ts": 837, - "src/neural/blender-adapter.test.ts": 828, - "src/cli/tune-cmd.test.ts": 826, - "src/testing/texture.test.ts": 815, - "src/render/input-image.test.ts": 812, - "src/material/generator.test.ts": 802, - "src/render/path/path-hero.test.ts": 782, - "src/audio/impact.test.ts": 760, - "src/render/icon.test.ts": 749, - "src/kinds/body-plan-asset.test.ts": 742, - "src/kinds/creature-clips.test.ts": 727, - "src/material/nature.test.ts": 725, - "src/cli/blender-session-cmd.test.ts": 670, - "src/render/review.test.ts": 631, - "src/passes/weather.test.ts": 602, - "src/qa/checks.test.ts": 527, - "src/material/strands.test.ts": 511, - "src/export/texture-sets.test.ts": 507, - "src/passes/belt-run.test.ts": 496, - "src/material/shader.test.ts": 478, - "src/physics/hull.test.ts": 478, - "src/finish/http.test.ts": 462, - "src/primitives/vdm.test.ts": 453, - "src/passes/simplify.test.ts": 450, - "scripts/parts-catalogue.test.ts": 439, - "src/qa/ref-fingerprint.test.ts": 411, - "src/qa/connections-held.test.ts": 409, - "src/render/blueprint.test.ts": 406, - "src/export/pivots-glb.test.ts": 398, - "src/audio/mechanism.test.ts": 397, - "src/qa/fur-silhouette.test.ts": 356, - "src/qa/crossing-contact.test.ts": 336, - "src/kinds/openings.test.ts": 332, - "src/kinds/body-plan-footlock.test.ts": 327, - "src/testing/connections.test.ts": 317, - "src/qa/chain-cross.test.ts": 316, - "src/passes/skin-heat.test.ts": 310, - "src/primitives/sdf.test.ts": 309, - "src/render/beauty-pbr.test.ts": 292, - "src/qa/connections.test.ts": 285, - "src/tune/objectives.test.ts": 285, - "src/tune/tune.test.ts": 281, - "src/finish/neural-texture.test.ts": 269, - "src/export/transmission.test.ts": 267, - "src/render/resolve.test.ts": 264, - "apps/site/src/glb-export.test.ts": 254, - "scripts/parts-catalogue-parts.test.ts": 254, - "src/export/pbr-validator.test.ts": 254, - "src/material/define.test.ts": 254, - "src/render/camera-solve.test.ts": 254, - "src/export/profile.test.ts": 247, - "src/export/textures.test.ts": 244, - "src/kinds/body-plan-weld.test.ts": 240, - "src/qa/deform.test.ts": 238, - "src/qa/gates.test.ts": 232, - "src/export/validator.test.ts": 230, - "src/passes/hair-roots.test.ts": 229, - "src/neural/blender-adapter-steps.test.ts": 228, - "src/passes/weather-paint.test.ts": 228, - "src/qa/finding-codes.test.ts": 226, - "src/render/beauty-cutout.test.ts": 225, - "src/export/tread-morphs.test.ts": 213, - "src/material/tex/lic.test.ts": 201, - "scripts/ref.test.ts": 199, - "scripts/public-tree.test.ts": 197, - "scripts/golden.test.ts": 194, - "scripts/vdm-import.test.ts": 190, - "src/kinds/hull.test.ts": 187, - "src/material/library.test.ts": 180, - "src/material/paint/fabric.test.ts": 178, - "src/passes/ray-grid.test.ts": 173, - "src/qa/swim.test.ts": 173, - "src/qa/ref-from-concept.test.ts": 171, - "src/passes/hair-cards.test.ts": 163, + "src/kinds/creature-family.test.ts": 29849, + "src/kinds/vegetation.test.ts": 18276, + "src/kinds/model-body.test.ts": 13324, + "src/finish/hero.test.ts": 10128, + "src/cli/bin.test.ts": 9895, + "src/primitives/sculpt-mesh-field.test.ts": 8815, + "src/cli/run.test.ts": 8384, + "scripts/ci-parity.test.ts": 8170, + "src/kinds/creature-types.test.ts": 7143, + "src/testing/readability.test.ts": 6757, + "examples/build-with-bun.test.ts": 6512, + "src/kinds/model-chain-two-pass.test.ts": 5937, + "src/testing/tiger.example.test.ts": 4917, + "src/kinds/model-rig.test.ts": 4698, + "src/cli/finish-cmd.test.ts": 4399, + "src/cli/mcp/server.test.ts": 4344, + "src/render/water-floor.test.ts": 4311, + "src/cli/tune-cmd.test.ts": 3947, + "src/render/beauty.test.ts": 3928, + "src/cli/render-cmds.test.ts": 3824, + "src/cli/concept-cmd.test.ts": 3767, + "src/qa/connections-seam.test.ts": 3680, + "src/kinds/model-fit.test.ts": 3451, + "src/kinds/creature-fur.test.ts": 3214, + "src/kinds/head.test.ts": 3140, + "src/kinds/creature.test.ts": 3007, + "src/cli/critique-cmd.test.ts": 2991, + "src/material/shader.test.ts": 2792, + "src/kinds/model-repose.test.ts": 2609, + "src/cli/run-json.test.ts": 2434, + "src/cli/compile-cache.test.ts": 2419, + "scripts/material-rows.test.ts": 2341, + "src/cli/mcp/tools.test.ts": 2322, + "src/neural/blender-session-lifecycle.test.ts": 2293, + "src/testing/assert.test.ts": 2150, + "src/cli/icon-cmd.test.ts": 2035, + "src/render/photo-match.test.ts": 1855, + "src/finish/bake.test.ts": 1738, + "scripts/verify.test.ts": 1596, + "src/render/set-shade.test.ts": 1543, + "src/render/path/path-hero.test.ts": 1473, + "src/cli/queue/contend.test.ts": 1457, + "src/neural/meshy-post.test.ts": 1399, + "src/audio/engine.test.ts": 1394, + "src/kinds/model-skeleton.test.ts": 1394, + "src/qa/gaps.test.ts": 1373, + "scripts/readme-example.test.ts": 1285, + "src/qa/connections-grown.test.ts": 1261, + "src/neural/tool-reap.test.ts": 1258, + "src/finish/finish.test.ts": 1230, + "src/audio/impact.test.ts": 1220, + "src/neural/blender-session-lock.test.ts": 1120, + "src/neural/blender-session.test.ts": 1120, + "scripts/lib/neural-cache.test.ts": 1048, + "src/render/input-image.test.ts": 1040, + "src/kinds/tentacle-clips.test.ts": 988, + "src/primitives/sculpt-stroke.test.ts": 986, + "src/primitives/sdf-detail.test.ts": 972, + "src/qa/critique-calibrate.test.ts": 972, + "src/kinds/body-plan-asset.test.ts": 900, + "src/finish/glb.test.ts": 890, + "src/cli/neural-cmd.test.ts": 820, + "src/qa/ref-gate.test.ts": 799, + "src/cli/neural-input.test.ts": 791, + "src/qa/critique-ref.test.ts": 790, + "src/render/icon.test.ts": 783, + "src/material/live-sheet.test.ts": 776, + "src/passes/weather.test.ts": 763, + "src/qa/chain-fft.test.ts": 740, + "src/kinds/creature-clips.test.ts": 722, + "src/kinds/roof.test.ts": 718, + "scripts/parts-catalogue.test.ts": 717, + "src/kinds/neck-clips.test.ts": 682, + "src/export/texture-sets.test.ts": 659, + "src/material/nature.test.ts": 631, + "src/material/generator.test.ts": 618, + "src/kinds/swim-clips.test.ts": 613, + "src/neural/blender-adapter.test.ts": 590, + "src/render/blueprint.test.ts": 580, + "src/cli/setup-cmd.test.ts": 575, + "src/physics/hull.test.ts": 572, + "src/primitives/vdm.test.ts": 548, + "src/audio/mechanism.test.ts": 534, + "src/export/pivots-glb.test.ts": 515, + "src/testing/connections.test.ts": 492, + "scripts/ref.test.ts": 486, + "src/testing/texture.test.ts": 474, + "src/finish/http.test.ts": 449, + "src/cli/blender-session-cmd.test.ts": 429, + "src/kinds/openings.test.ts": 429, + "src/passes/belt-run.test.ts": 420, + "src/render/review.test.ts": 417, + "src/passes/simplify.test.ts": 385, + "src/material/strands.test.ts": 383, + "apps/site/src/glb-export.test.ts": 375, + "scripts/parts-catalogue-parts.test.ts": 370, + "src/kinds/hull.test.ts": 364, + "src/render/beauty-cutout.test.ts": 364, + "src/export/pbr-validator.test.ts": 361, + "src/passes/skin-heat.test.ts": 360, + "src/qa/crossing-contact.test.ts": 358, + "src/qa/checks.test.ts": 352, + "src/render/camera-solve.test.ts": 341, + "src/qa/ref-fingerprint.test.ts": 314, + "src/export/validator.test.ts": 303, + "src/kinds/body-plan-footlock.test.ts": 302, + "src/cli/queue/cmd.test.ts": 296, + "src/export/transmission.test.ts": 296, + "src/passes/hair-roots.test.ts": 294, + "src/neural/blender-adapter-steps.test.ts": 291, + "src/export/tread-morphs.test.ts": 285, + "src/material/define.test.ts": 271, + "src/qa/connections.test.ts": 261, + "scripts/public-tree.test.ts": 257, + "src/kinds/groom-guides.test.ts": 242, + "src/primitives/sdf.test.ts": 235, + "src/finish/neural-texture.test.ts": 224, + "src/testing/gaps.test.ts": 217, + "src/passes/weather-paint.test.ts": 213, + "src/qa/deform.test.ts": 199, + "src/material/tex/lic.test.ts": 192, + "src/render/beauty-pbr.test.ts": 190, + "src/qa/finding-codes.test.ts": 178, + "src/render/path/bvh.test.ts": 177, + "src/qa/connections-held.test.ts": 163, "src/primitives/sculpt-tree.test.ts": 162, - "src/testing/gaps.test.ts": 162, - "scripts/release.test.ts": 161, - "src/neural/meshy-adapter.test.ts": 157, - "src/finish/lscm.test.ts": 155, - "src/material/paint/metal-values.test.ts": 149, - "src/kinds/groom-guides.test.ts": 142, - "src/qa/critique-vector.test.ts": 141, - "src/kinds/creature-families.test.ts": 136, - "src/primitives/sculpt.test.ts": 129, - "src/cli/build-one.test.ts": 125, - "src/render/path/hair.test.ts": 125, - "src/kinds/creature-quadruped.test.ts": 123, - "src/render/path/bvh.test.ts": 123, - "src/qa/clash.test.ts": 122, - "src/data/terrarium.test.ts": 115, - "src/physics/collider.test.ts": 106, - "src/cli/blender-review-cmd.test.ts": 103, - "scripts/fetch-timeout.test.ts": 102, - "src/finish/split-parts.test.ts": 101, - "src/qa/treads.test.ts": 100, - "src/neural/http-adapter.test.ts": 98, - "src/neural/refine.test.ts": 98, - "src/primitives/sdf-silhouette.test.ts": 98, - "src/material/paint/plates.test.ts": 92, - "src/passes/imperfect.test.ts": 92, - "src/render/blueprint-legend.test.ts": 91, - "src/qa/critique-shape.test.ts": 88, - "src/neural/meshy-rig.test.ts": 87, - "src/passes/bake-field.test.ts": 84, - "src/cli/manifest.test.ts": 74, - "src/core/deflate.test.ts": 74, - "src/neural/import-glb.test.ts": 73, - "src/passes/bake-normals.test.ts": 73, - "src/render/beauty.test.ts": 72, - "src/passes/hair-cards-shape.test.ts": 71, - "src/primitives/vdm-scatter.test.ts": 70, - "src/export/physics.test.ts": 64, - "src/qa/clash-path.test.ts": 62, - "src/primitives/terrain.test.ts": 61, - "src/qa/clash-thin.test.ts": 59, - "src/kinds/model-chain.test.ts": 56, - "src/qa/coherence.test.ts": 55, - "scripts/verify.test.ts": 54, - "src/kinds/body-plan-hooks.test.ts": 53, - "src/kinds/model-bind.test.ts": 53, - "src/qa/floating.test.ts": 53, - "src/core/png.test.ts": 52, - "src/export/skin.test.ts": 52, - "src/cli/neural-spend.test.ts": 50, - "src/neural/kubeez-adapter.test.ts": 50, - "src/testing/image.test.ts": 50, - "src/qa/model.test.ts": 48, - "src/render/sheet.test.ts": 48, - "src/export/draw-budget.test.ts": 46, - "src/kinds/gait.test.ts": 46, - "src/render/path/bsdf.test.ts": 43, - "src/kinds/model-fit-joints.test.ts": 42, - "src/physics/occupancy.test.ts": 41, - "src/material/tex/graph.test.ts": 40, - "src/kinds/model-joints.test.ts": 39, - "apps/site/src/texture-sets.test.ts": 38, - "src/passes/merge-geometry.test.ts": 38, - "src/render/metrics.test.ts": 38, - "src/passes/fair.test.ts": 36, - "src/kinds/city.test.ts": 35, - "src/primitives/terrain-grid.test.ts": 35, - "src/primitives/text-vector.test.ts": 32, - "src/primitives/vdm-library.test.ts": 32, - "src/qa/connections-water.test.ts": 32, - "src/cli/config.test.ts": 31, - "src/kinds/chain-bake.test.ts": 31, - "src/primitives/svg-extrude.test.ts": 31, - "src/qa/clash-spin.test.ts": 31, - "src/qa/gait.test.ts": 30, - "scripts/data.test.ts": 29, - "src/passes/bake-height.test.ts": 29, - "src/passes/weather-select.test.ts": 29, - "src/qa/measures.test.ts": 29, - "src/material/slots.test.ts": 28, - "src/cli/mcp/images.test.ts": 27, - "src/neural/png-read.test.ts": 27, - "src/primitives/sdf-ops.test.ts": 27, - "src/cli/watch.test.ts": 26, - "src/material/variant.test.ts": 25, - "src/qa/model-doubled.test.ts": 24, - "src/testing/mesh.test.ts": 23, - "src/export/tread-speed.test.ts": 22, - "src/finish/maps.test.ts": 22, - "src/passes/hair-flow.test.ts": 22, - "src/physics/mass.test.ts": 22, + "src/qa/fur-silhouette.test.ts": 161, + "src/render/path/hair.test.ts": 158, + "src/cli/queue/tickets.test.ts": 154, + "src/material/paint/fabric.test.ts": 152, + "src/passes/hair-cards.test.ts": 148, + "src/passes/ray-grid.test.ts": 148, + "src/kinds/creature-quadruped.test.ts": 147, + "src/physics/collider.test.ts": 144, + "src/neural/meshy-adapter.test.ts": 139, + "src/primitives/sculpt.test.ts": 135, + "src/export/physics.test.ts": 134, + "src/data/terrarium.test.ts": 130, + "src/cli/build-one.test.ts": 128, + "src/kinds/creature-families.test.ts": 124, + "src/kinds/body-plan-weld.test.ts": 120, + "src/passes/bake-field.test.ts": 113, + "scripts/fetch-timeout.test.ts": 111, + "src/export/profile.test.ts": 108, + "src/material/paint/library-relief.test.ts": 107, + "src/material/paint/plates.test.ts": 106, + "src/neural/http-adapter.test.ts": 106, + "src/render/resolve.test.ts": 105, + "src/export/textures.test.ts": 103, + "src/qa/swim.test.ts": 102, + "src/primitives/terrain.test.ts": 100, + "src/qa/chain-cross.test.ts": 99, + "src/qa/clash.test.ts": 98, + "src/tune/tune.test.ts": 98, + "src/kinds/model-bind.test.ts": 97, + "src/finish/split-parts.test.ts": 95, + "src/cli/queue/acquire.test.ts": 94, + "src/render/sheet.test.ts": 94, + "src/export/draw-budget.test.ts": 93, + "src/tune/objectives.test.ts": 90, + "scripts/release.test.ts": 89, + "src/passes/imperfect.test.ts": 89, + "src/qa/gates.test.ts": 88, + "src/cli/manifest.test.ts": 83, + "src/finish/lscm.test.ts": 80, + "src/core/deflate.test.ts": 78, + "src/passes/hair-cards-shape.test.ts": 77, + "src/testing/image.test.ts": 76, + "src/core/png.test.ts": 73, + "src/qa/ref-from-concept.test.ts": 70, + "src/primitives/vdm-scatter.test.ts": 69, + "src/qa/treads.test.ts": 68, + "src/export/skin.test.ts": 67, + "src/render/path/bsdf.test.ts": 66, + "src/passes/bake-normals.test.ts": 65, + "src/primitives/sdf-silhouette.test.ts": 63, + "src/neural/refine.test.ts": 62, + "apps/site/src/texture-sets.test.ts": 61, + "src/cli/mcp/images.test.ts": 60, + "src/export/clips-lod.test.ts": 56, + "src/passes/merge-geometry.test.ts": 54, + "src/qa/critique-vector.test.ts": 51, + "src/qa/model.test.ts": 51, + "src/physics/occupancy.test.ts": 49, + "scripts/data.test.ts": 45, + "src/neural/meshy-rig.test.ts": 45, + "src/material/tex/graph.test.ts": 43, + "src/primitives/vdm-library.test.ts": 40, + "src/cli/queue/fifo.test.ts": 39, + "src/testing/mesh.test.ts": 39, + "src/export/lod-glb.test.ts": 38, + "src/kinds/creature-pattern.test.ts": 38, + "src/passes/fair.test.ts": 38, + "src/qa/chain-twist.test.ts": 38, + "src/primitives/text-vector.test.ts": 37, + "src/qa/connections-water.test.ts": 37, + "src/kinds/body-plan-hooks.test.ts": 36, + "src/neural/kubeez-adapter.test.ts": 36, + "scripts/vdm-from-blender.test.ts": 35, + "src/primitives/svg-extrude.test.ts": 35, + "src/passes/bake-height.test.ts": 34, + "src/primitives/sdf-ops.test.ts": 33, + "src/render/blueprint-legend.test.ts": 33, + "src/qa/clash-thin.test.ts": 32, + "src/data/terrain-from-dem.test.ts": 31, + "src/export/clips.test.ts": 31, + "src/qa/clash-path.test.ts": 31, + "src/qa/clash-spin.test.ts": 30, + "scripts/golden.test.ts": 29, + "src/cli/blender-review-cmd.test.ts": 29, + "src/export/tread-speed.test.ts": 28, + "src/kinds/chain-bake.test.ts": 27, + "src/primitives/terrain-grid.test.ts": 27, + "src/render/pose-strip.test.ts": 27, + "src/material/variant.test.ts": 26, + "src/kinds/groom-ops.test.ts": 25, + "src/material/slots.test.ts": 25, + "src/qa/clip-pose.test.ts": 24, + "src/export/audio.test.ts": 22, + "src/kinds/model-chain.test.ts": 22, + "src/neural/import-model.test.ts": 22, + "src/neural/tool-adapters.test.ts": 22, "src/qa/skin.test.ts": 22, - "src/render/silhouette-sheet.test.ts": 22, - "src/material/tex/stamps-organic.test.ts": 21, - "src/qa/clip-pose.test.ts": 21, - "src/qa/joints.test.ts": 21, - "src/data/terrain-from-dem.test.ts": 20, - "src/kinds/creature-pattern.test.ts": 20, - "src/kinds/head-clips.test.ts": 20, - "src/qa/measure.test.ts": 20, - "src/export/audio.test.ts": 19, - "src/neural/tool-adapters.test.ts": 19, - "src/passes/weather-subdivide.test.ts": 19, - "scripts/vdm-from-blender.test.ts": 18, - "src/neural/import-model.test.ts": 18, - "src/neural/recipe-cache.test.ts": 18, - "src/primitives/lathe.test.ts": 18, - "src/qa/chain-twist.test.ts": 18, - "src/render/texture-sets.test.ts": 18, - "src/export/lod-glb.test.ts": 17, - "src/kinds/groom-cards.test.ts": 17, - "src/neural/godot-adapter.test.ts": 17, - "src/neural/import-clips.test.ts": 17, - "src/material/paint/image-strip.test.ts": 16, - "src/neural/blender-step-stroke.test.ts": 16, - "src/tune/cmaes.test.ts": 16, - "src/neural/blender-step-order.test.ts": 15, - "src/passes/skin-weights.test.ts": 15, - "src/qa/check-selectors.test.ts": 15, - "src/render/path/denoise.test.ts": 15, - "src/render/thumbnail.test.ts": 15, - "src/testing/measure.test.ts": 15, - "src/cli/doctor-cmd.test.ts": 14, - "src/kinds/body-plan.test.ts": 14, - "src/export/clips-lod.test.ts": 13, - "src/kinds/groom-ops.test.ts": 13, - "src/kinds/neural-hair.test.ts": 13, - "src/passes/optimize.test.ts": 13, + "src/render/path/denoise.test.ts": 22, + "src/physics/mass.test.ts": 21, + "src/cli/neural-spend.test.ts": 20, + "src/passes/skin-weights.test.ts": 20, + "src/material/tex/stamps-organic.test.ts": 19, + "src/neural/recipe-cache.test.ts": 19, + "src/passes/hair-flow.test.ts": 19, + "src/render/path/sss.test.ts": 19, + "src/finish/maps.test.ts": 18, + "src/kinds/city.test.ts": 18, + "src/render/thumbnail.test.ts": 18, + "scripts/lib/shard.test.ts": 17, + "src/cli/doctor-cmd.test.ts": 17, + "src/export/clip-channels.test.ts": 17, + "src/kinds/head-clips.test.ts": 17, + "src/primitives/erosion-memo.test.ts": 16, + "src/qa/coherence.test.ts": 16, + "src/qa/critique-shape.test.ts": 16, + "src/render/texture-sets.test.ts": 16, + "src/data/sha256.test.ts": 15, + "src/primitives/triangulate.test.ts": 15, + "scripts/pack-budget.test.ts": 14, + "src/passes/optimize.test.ts": 14, + "src/cli/config.test.ts": 13, + "src/cli/run-json-errors.test.ts": 13, + "src/cli/watch.test.ts": 13, + "src/core/joint-fit.test.ts": 13, + "src/core/predicates.test.ts": 13, + "src/kinds/groom-cards.test.ts": 13, + "src/neural/godot-adapter.test.ts": 13, "src/primitives/lattice.test.ts": 13, - "src/cli/init-cmd.test.ts": 12, - "src/kinds/head-tone.test.ts": 12, - "src/qa/lods.test.ts": 12, - "src/qa/point-triangle.test.ts": 12, - "src/testing/stability.test.ts": 12, - "scripts/lib/critique.test.ts": 11, - "src/audio/wav.test.ts": 11, - "src/export/materials.test.ts": 11, - "src/physics/linalg.test.ts": 11, - "src/primitives/noise.test.ts": 11, - "src/render/path/sss.test.ts": 11, - "src/export/clips.test.ts": 10, - "src/finish/hero-qa.test.ts": 10, - "src/kinds/model-character.test.ts": 10, - "src/neural/import-skin.test.ts": 10, - "src/primitives/similarity-blend.test.ts": 10, - "scripts/textures.test.ts": 9, + "src/primitives/noise.test.ts": 13, + "src/qa/gait.test.ts": 13, + "src/export/own-textures.test.ts": 12, + "src/kinds/gait.test.ts": 12, + "src/kinds/neural-motion.test.ts": 12, + "src/passes/bake-ao.test.ts": 12, + "src/passes/weather-select.test.ts": 12, + "src/qa/joints.test.ts": 12, + "scripts/ci-changes.test.ts": 11, + "src/kinds/model-character.test.ts": 11, + "src/qa/point-triangle.test.ts": 11, + "src/render/alpha-mips.test.ts": 11, + "src/render/silhouette-sheet.test.ts": 11, + "src/testing/measure.test.ts": 11, + "src/tune/cmaes.test.ts": 11, + "scripts/vdm-import.test.ts": 10, + "src/export/variants.test.ts": 10, + "src/neural/adapter.test.ts": 10, + "src/qa/measure.test.ts": 10, + "src/testing/stability.test.ts": 10, "src/primitives/similarity-fit.test.ts": 9, + "src/testing/lod.test.ts": 9, + "apps/site/src/measure-overlay.test.ts": 8, "scripts/conventions.test.ts": 8, - "src/cli/args.test.ts": 8, - "src/core/limit-tolerance.test.ts": 8, - "src/core/predicates.test.ts": 8, - "src/data/sha256.test.ts": 8, - "src/export/pbr.test.ts": 8, - "src/export/variants.test.ts": 8, - "src/neural/meshy-task.test.ts": 8, - "src/passes/bake-ao.test.ts": 8, - "src/qa/scene-budget.test.ts": 8, + "src/export/morph-targets.test.ts": 8, + "src/finish/hero-detail.test.ts": 8, + "src/neural/import-clips.test.ts": 8, + "src/neural/import-glb.test.ts": 8, + "src/neural/png-read.test.ts": 8, + "src/primitives/winding.test.ts": 8, + "src/qa/floating.test.ts": 8, + "src/qa/skin-sides.test.ts": 8, "src/render/brdf.test.ts": 8, - "src/testing/lod.test.ts": 8, - "scripts/fonts-hershey.test.ts": 7, - "src/cli/load.test.ts": 7, - "src/core/validate-asset.test.ts": 7, - "src/kinds/model-character-chain.test.ts": 7, - "src/material/paint/weather.test.ts": 7, - "src/neural/comfy-adapter.test.ts": 7, - "src/primitives/erosion-memo.test.ts": 7, + "scripts/lib/critique.test.ts": 7, + "src/cli/queue/doctor.test.ts": 7, + "src/export/glb.test.ts": 7, + "src/export/merge-pivots.test.ts": 7, + "src/export/pbr.test.ts": 7, + "src/kinds/body-plan.test.ts": 7, + "src/kinds/creature-tone.test.ts": 7, + "src/kinds/model-fit-joints.test.ts": 7, + "src/passes/weather-subdivide.test.ts": 7, "src/primitives/text.test.ts": 7, - "src/primitives/triangulate.test.ts": 7, - "src/qa/critique.test.ts": 7, "scripts/changelog.test.ts": 6, - "scripts/data-osm.test.ts": 6, - "scripts/site-shots.test.ts": 6, - "src/cli/mcp/explain.test.ts": 6, - "src/cli/plan-template.test.ts": 6, - "src/cli/run-json-errors.test.ts": 6, - "src/export/own-textures.test.ts": 6, - "src/kinds/radial-pulse.test.ts": 6, + "src/cli/init-cmd.test.ts": 6, + "src/export/glb-inputs.test.ts": 6, + "src/export/instancing.test.ts": 6, + "src/kinds/model-joints.test.ts": 6, "src/kinds/springs.test.ts": 6, - "src/kinds/vehicle.test.ts": 6, - "src/neural/sha256.test.ts": 6, - "src/neural/wire.test.ts": 6, - "src/passes/weighted-normals.test.ts": 6, - "src/primitives/loft.test.ts": 6, - "src/primitives/prism.test.ts": 6, - "src/primitives/warp.test.ts": 6, - "src/primitives/winding.test.ts": 6, - "src/qa/occlusion.test.ts": 6, - "apps/site/src/measure-overlay.test.ts": 5, - "apps/site/src/scenarios/swim.test.ts": 5, - "src/export/clip-channels.test.ts": 5, - "src/export/glb.test.ts": 5, - "src/export/instancing.test.ts": 5, - "src/kinds/creature-tone.test.ts": 5, - "src/kinds/mass.test.ts": 5, - "src/neural/meshy-net.test.ts": 5, - "src/neural/openai-image-adapter.test.ts": 5, - "src/qa/segment-contact.test.ts": 5, - "src/render/alpha-mips.test.ts": 5, - "src/render/part-masks.test.ts": 5, - "src/core/joint-fit.test.ts": 4, + "src/neural/import-skin.test.ts": 6, + "src/qa/check-selectors.test.ts": 6, + "src/qa/segment-contact.test.ts": 6, + "scripts/textures.test.ts": 5, + "src/cli/args.test.ts": 5, + "src/core/define-part.test.ts": 5, + "src/core/validate-asset.test.ts": 5, + "src/material/paint/image-strip.test.ts": 5, + "src/neural/sha256.test.ts": 5, + "src/primitives/box.test.ts": 5, + "src/primitives/sweep.test.ts": 5, + "src/primitives/warp.test.ts": 5, + "src/qa/scene-budget.test.ts": 5, + "src/testing/poly.test.ts": 5, + "apps/site/src/credits-panel.test.ts": 4, + "scripts/data-osm.test.ts": 4, + "scripts/shots.test.ts": 4, + "scripts/textures-zip.test.ts": 4, + "src/cli/mcp/explain.test.ts": 4, + "src/cli/server-cmd.test.ts": 4, "src/data/dem.test.ts": 4, - "src/export/glb-inputs.test.ts": 4, - "src/export/merge-pivots.test.ts": 4, - "src/export/morph-targets.test.ts": 4, - "src/kinds/neural-motion.test.ts": 4, - "src/neural/blender-steps.test.ts": 4, - "src/passes/morph.test.ts": 4, - "src/render/font.test.ts": 4, - "src/testing/poly.test.ts": 4, - "apps/site/src/credits-panel.test.ts": 3, + "src/export/motion.test.ts": 4, + "src/export/skin-quantise.test.ts": 4, + "src/kinds/mass.test.ts": 4, + "src/physics/check.test.ts": 4, + "src/primitives/erosion.test.ts": 4, + "src/primitives/lathe.test.ts": 4, + "src/primitives/loft.test.ts": 4, + "src/qa/measures.test.ts": 4, "apps/site/src/scenarios/tread-belt.test.ts": 3, + "scripts/lib/files.test.ts": 3, "scripts/recipes.test.ts": 3, - "scripts/shots.test.ts": 3, - "src/data/real-from.test.ts": 3, - "src/neural/adapter.test.ts": 3, - "src/neural/import-clips-loop.test.ts": 3, - "src/neural/record.test.ts": 3, - "src/passes/bake-texels.test.ts": 3, - "src/passes/livery.test.ts": 3, - "src/physics/density.test.ts": 3, + "src/cli/plan-template.test.ts": 3, + "src/core/asset.test.ts": 3, + "src/core/finish.test.ts": 3, + "src/core/motion-pose.test.ts": 3, + "src/core/neural-task.test.ts": 3, + "src/data/osm.test.ts": 3, + "src/export/node-names.test.ts": 3, + "src/kinds/creature-fur-look.test.ts": 3, + "src/kinds/creature-muscles.test.ts": 3, + "src/kinds/model-character-chain.test.ts": 3, + "src/kinds/vehicle.test.ts": 3, + "src/material/creature.test.ts": 3, + "src/material/paint/metal-values.test.ts": 3, + "src/neural/blender-steps.test.ts": 3, + "src/neural/comfy-adapter.test.ts": 3, + "src/neural/generator.test.ts": 3, + "src/neural/import-clean.test.ts": 3, + "src/neural/meshy-task.test.ts": 3, + "src/neural/openai-image-adapter.test.ts": 3, + "src/passes/morph.test.ts": 3, "src/primitives/svg-path.test.ts": 3, - "src/primitives/sweep.test.ts": 3, - "src/qa/skin-sides.test.ts": 3, + "src/qa/critique.test.ts": 3, + "src/qa/model-doubled.test.ts": 3, + "src/qa/occlusion.test.ts": 3, + "src/qa/sanity.test.ts": 3, + "src/render/metrics.test.ts": 3, "src/testing/follows.test.ts": 3, "apps/site/src/scenarios/clips.test.ts": 2, "apps/site/src/worker-pool.test.ts": 2, "scripts/boundaries.test.ts": 2, - "scripts/determinism.test.ts": 2, + "scripts/golden-brief.test.ts": 2, "scripts/import.test.ts": 2, - "scripts/lib/files.test.ts": 2, "scripts/lib/neural-licence.test.ts": 2, - "scripts/pack-smoke.test.ts": 2, - "src/cli/index-json.test.ts": 2, - "src/cli/mcp/sandbox.test.ts": 2, + "src/cli/load.test.ts": 2, + "src/cli/queue/config.test.ts": 2, + "src/cli/queue/liveness.test.ts": 2, "src/core/asset-use-skin.test.ts": 2, - "src/core/asset.test.ts": 2, "src/core/clip.test.ts": 2, - "src/core/define-part.test.ts": 2, - "src/core/finish.test.ts": 2, "src/core/paint.test.ts": 2, "src/core/skeleton.test.ts": 2, "src/core/texture-set.test.ts": 2, "src/data/wikidata.test.ts": 2, - "src/export/credits.test.ts": 2, - "src/export/motion.test.ts": 2, - "src/export/node-names.test.ts": 2, + "src/export/materials.test.ts": 2, "src/export/tangents.test.ts": 2, - "src/finish/hero-detail.test.ts": 2, "src/kinds/chain-drivers.test.ts": 2, - "src/kinds/chain-field.test.ts": 2, - "src/kinds/creature-muscles.test.ts": 2, - "src/material/creature.test.ts": 2, + "src/kinds/creature-bones.test.ts": 2, + "src/kinds/neural-hair.test.ts": 2, + "src/kinds/vegetation-needles.test.ts": 2, "src/material/paint/relief.test.ts": 2, - "src/neural/generator.test.ts": 2, - "src/neural/import-clean.test.ts": 2, + "src/neural/blender-step-order.test.ts": 2, + "src/neural/blender-step-stroke.test.ts": 2, + "src/neural/record.test.ts": 2, + "src/passes/weighted-normals.test.ts": 2, "src/primitives/beam.test.ts": 2, - "src/primitives/box.test.ts": 2, - "src/primitives/erosion.test.ts": 2, + "src/primitives/check.test.ts": 2, "src/primitives/heightfield.test.ts": 2, - "src/qa/sanity.test.ts": 2, - "src/qa/weather.test.ts": 2, - "src/render/lighting.test.ts": 2, - "apps/site/src/autoplay.test.ts": 1, - "apps/site/src/categories.test.ts": 1, - "apps/site/src/flagships.test.ts": 1, - "apps/site/src/route.test.ts": 1, - "apps/site/src/shown.test.ts": 1, - "apps/site/src/sphere-fit.test.ts": 1, + "src/qa/lods.test.ts": 2, + "src/render/part-masks.test.ts": 2, + "src/render/path/sky.test.ts": 2, + "apps/site/src/compare.test.ts": 1, + "apps/site/src/scenarios/swim.test.ts": 1, "apps/site/src/tier-toggle.test.ts": 1, "apps/site/src/viewer-fallback.test.ts": 1, "scripts/conventions-brand.test.ts": 1, - "scripts/golden-brief.test.ts": 1, - "scripts/lib/bench-table.test.ts": 1, + "scripts/determinism.test.ts": 1, + "scripts/fonts-hershey.test.ts": 1, + "scripts/pack-smoke.test.ts": 1, "scripts/placeholders.test.ts": 1, + "scripts/site-shots.test.ts": 1, + "src/cli/concept-cli.test.ts": 1, + "src/cli/mcp/sandbox.test.ts": 1, + "src/cli/neural-store.test.ts": 1, + "src/cli/queue/needs.test.ts": 1, "src/cli/report.test.ts": 1, "src/core/family.test.ts": 1, "src/core/mesh.test.ts": 1, "src/core/motion-check.test.ts": 1, - "src/core/motion-pose.test.ts": 1, "src/core/neural-ctx.test.ts": 1, - "src/core/neural-task.test.ts": 1, - "src/core/optics.test.ts": 1, + "src/core/style.test.ts": 1, "src/core/transform.test.ts": 1, "src/core/tune.test.ts": 1, - "src/data/osm.test.ts": 1, - "src/export/hierarchy.test.ts": 1, + "src/data/real-from.test.ts": 1, + "src/export/credits.test.ts": 1, "src/export/lights.test.ts": 1, - "src/export/version.test.ts": 1, "src/grammar/split.test.ts": 1, "src/kinds/building.test.ts": 1, + "src/kinds/chain-field.test.ts": 1, + "src/kinds/head-tone.test.ts": 1, + "src/kinds/radial-pulse.test.ts": 1, "src/kinds/tread.test.ts": 1, - "src/kinds/vegetation-needles.test.ts": 1, - "src/material/material-def.test.ts": 1, + "src/material/library.test.ts": 1, "src/material/paint/stamp.test.ts": 1, - "src/material/paint/validate.test.ts": 1, - "src/material/trim-sheet.test.ts": 1, "src/neural/blender-gpu.test.ts": 1, - "src/neural/blender-scripts.test.ts": 1, "src/neural/blender-skeleton.test.ts": 1, "src/neural/glb-read.test.ts": 1, + "src/neural/import-clips-loop.test.ts": 1, + "src/neural/meshy-net.test.ts": 1, "src/neural/meshy-params.test.ts": 1, - "src/neural/png-fit.test.ts": 1, - "src/passes/grime.test.ts": 1, - "src/physics/check.test.ts": 1, + "src/neural/wire.test.ts": 1, + "src/passes/bake-texels.test.ts": 1, + "src/physics/linalg.test.ts": 1, "src/primitives/cell.test.ts": 1, - "src/primitives/check.test.ts": 1, "src/primitives/planar.test.ts": 1, + "src/primitives/prism.test.ts": 1, + "src/primitives/similarity-blend.test.ts": 1, "src/primitives/tiled.test.ts": 1, "src/qa/geometry-cache.test.ts": 1, - "src/qa/geometry.test.ts": 1, "src/qa/mass.test.ts": 1, + "src/qa/weather.test.ts": 1, + "src/render/font.test.ts": 1, + "src/render/lighting.test.ts": 1, "src/render/options.test.ts": 1, "src/render/path/classes.test.ts": 1, - "src/render/path/sky.test.ts": 1, + "src/render/path/parallel.test.ts": 1, + "src/render/path/sampler.test.ts": 1, "src/render/review-warnings.test.ts": 1, - "src/render/vis-buffer.test.ts": 1, - "apps/site/src/compare.test.ts": 0, + "apps/site/src/autoplay.test.ts": 0, + "apps/site/src/categories.test.ts": 0, + "apps/site/src/flagships.test.ts": 0, "apps/site/src/lod-policy.test.ts": 0, "apps/site/src/retry.test.ts": 0, + "apps/site/src/route.test.ts": 0, + "apps/site/src/shown.test.ts": 0, + "apps/site/src/sphere-fit.test.ts": 0, "scripts/build.test.ts": 0, + "scripts/lib/bench-table.test.ts": 0, "scripts/lib/recipe-imports.test.ts": 0, - "scripts/lib/shard.test.ts": 0, - "scripts/textures-zip.test.ts": 0, - "src/cli/concept-cli.test.ts": 0, + "scripts/recipe-costs.test.ts": 0, + "src/audio/wav.test.ts": 0, + "src/cli/index-json.test.ts": 0, "src/core/errors.test.ts": 0, "src/core/format-count.test.ts": 0, + "src/core/limit-tolerance.test.ts": 0, "src/core/medium.test.ts": 0, + "src/core/optics.test.ts": 0, "src/core/rng.test.ts": 0, - "src/core/style.test.ts": 0, - "src/export/skin-quantise.test.ts": 0, + "src/export/hierarchy.test.ts": 0, + "src/export/version.test.ts": 0, + "src/finish/hero-qa.test.ts": 0, "src/grammar/whyte.test.ts": 0, - "src/kinds/creature-bones.test.ts": 0, + "src/material/material-def.test.ts": 0, + "src/material/paint/validate.test.ts": 0, + "src/material/paint/weather.test.ts": 0, "src/material/palette.test.ts": 0, + "src/material/trim-sheet.test.ts": 0, + "src/neural/blender-scripts.test.ts": 0, + "src/neural/png-fit.test.ts": 0, "src/passes/belt-seam.test.ts": 0, - "src/render/path/parallel.test.ts": 0, - "src/render/path/sampler.test.ts": 0 + "src/passes/grime.test.ts": 0, + "src/passes/livery.test.ts": 0, + "src/physics/density.test.ts": 0, + "src/qa/geometry.test.ts": 0, + "src/render/vis-buffer.test.ts": 0 } } diff --git a/scripts/pack-budget.test.ts b/scripts/pack-budget.test.ts new file mode 100644 index 00000000..6001c7de --- /dev/null +++ b/scripts/pack-budget.test.ts @@ -0,0 +1,83 @@ +// The tarball's weight limits: a listing read from real tar bytes, and every over-budget case named. + +import { expect, it } from 'bun:test'; +import { mkdir, mkdtemp, rm, writeFile } from 'node:fs/promises'; +import { tmpdir } from 'node:os'; +import { dirname, join } from 'node:path'; +import { budgetProblems, PACK_BUDGET, type TarEntry, tarEntries } from './pack-budget.ts'; + +const MB = 1_000_000; +const LONG = `src/${'deep/'.repeat(25)}file.ts`; + +async function tgz(format: 'gnu' | 'pax', files: Record): Promise { + const dir = await mkdtemp(join(tmpdir(), 'ts23d-pack-budget-')); + try { + for (const [path, size] of Object.entries(files)) { + const at = join(dir, 'package', path); + await mkdir(dirname(at), { recursive: true }); + await writeFile(at, new Uint8Array(size)); + } + const out = join(dir, 'p.tgz'); + const p = Bun.spawnSync(['tar', `--format=${format}`, '-czf', out, 'package'], { cwd: dir }); + if (p.exitCode !== 0) throw new Error(p.stderr.toString()); + return new Uint8Array(await Bun.file(out).arrayBuffer()); + } finally { + await rm(dir, { recursive: true, force: true }); + } +} + +const sorted = (e: readonly TarEntry[]) => [...e].sort((a, b) => (a.path < b.path ? -1 : 1)); + +for (const format of ['gnu', 'pax'] as const) { + it(`reads every regular file, its size and a > 100-char path from a ${format} tarball`, async () => { + const entries = tarEntries(await tgz(format, { 'package.json': 12, [LONG]: 700, 'a/b.md': 0 })); + expect(sorted(entries)).toEqual([ + { path: 'a/b.md', size: 0 }, + { path: 'package.json', size: 12 }, + { path: LONG, size: 700 }, + ]); + }); +} + +it('rejects a tarball over the packed or unpacked budget and names the largest files', () => { + const entries = [ + { path: 'dist/index.js', size: 0.5 * MB }, + { path: 'src/index.ts', size: 0.2 * MB }, + ]; + const budget = { packedBytes: 1 * MB, unpackedBytes: 0.6 * MB, fileBytes: 1 * MB }; + const problems = budgetProblems(entries, 2 * MB, budget); + expect(problems).toHaveLength(2); + expect(problems[0]).toStartWith('tarball is 2.00 MB packed, budget 1.00 MB'); + expect(problems[1]).toStartWith('tarball is 0.70 MB unpacked, budget 0.60 MB'); + expect(problems[1]).toContain('dist/index.js (0.50 MB), src/index.ts (0.20 MB)'); + expect(problems[1]).toContain('package.json "files"'); +}); + +it('rejects a single file over the per-file limit', () => { + const problems = budgetProblems([{ path: 'docs/media/demo.mp4', size: 3 * MB }], 1, { + ...PACK_BUDGET, + unpackedBytes: 10 * MB, + }); + expect(problems).toEqual([ + 'ships a 3.00 MB file (limit 1.00 MB): docs/media/demo.mp4; exclude it in package.json "files"', + ]); +}); + +it('rejects any recipes/*/neural/ file, however small: model outputs stay in git', () => { + const problems = budgetProblems( + [ + { path: 'recipes/bear_hybrid/neural/abc/record.json', size: 900 }, + { path: 'recipes/bear_hybrid/bear_hybrid.ts', size: 900 }, + ], + 1, + ); + expect(problems).toHaveLength(1); + expect(problems[0]).toStartWith( + 'ships a model output: recipes/bear_hybrid/neural/abc/record.json', + ); + expect(problems[0]).toContain('"!recipes/*/neural"'); +}); + +it('accepts a listing inside every limit', () => { + expect(budgetProblems([{ path: 'dist/index.js', size: 0.5 * MB }], 0.2 * MB)).toEqual([]); +}); diff --git a/scripts/pack-budget.ts b/scripts/pack-budget.ts new file mode 100644 index 00000000..1adea225 --- /dev/null +++ b/scripts/pack-budget.ts @@ -0,0 +1,111 @@ +// What the npm tarball may weigh and carry: a packed and unpacked size budget, a per-file limit and +// no `recipes/*/neural/` model outputs (they stay in git; without them `ranger` and `human_head` +// build their CPU tier, and the model-first hybrids stop with T3D_NEURAL_MISSING). `tarEntries` reads paths and sizes straight from the .tgz bytes (ustar, GNU +// long names, pax paths), so the check needs no `tar` flavour. Used by scripts/pack-smoke.ts. + +import { gunzipSync } from 'node:zlib'; + +export interface TarEntry { + /** Path inside the package, without the leading `package/`. */ + readonly path: string; + readonly size: number; +} + +export interface PackBudget { + readonly packedBytes: number; + readonly unpackedBytes: number; + readonly fileBytes: number; +} + +/** + * Measured 2026-11-19 with `recipes/*\/neural` excluded: 4.22 MB packed, 12.7 MB unpacked, largest + * file 0.50 MB (a dist chunk). Limits sit ~25 % above; raise them only with a reason in the diff. + */ +export const PACK_BUDGET: PackBudget = { + packedBytes: 5_300_000, + unpackedBytes: 16_000_000, + fileBytes: 1_000_000, +}; + +const NEURAL = /^recipes\/[^/]+\/neural\//; +const decoder = new TextDecoder(); + +function field(b: Uint8Array, at: number, len: number): string { + const s = b.subarray(at, at + len); + const end = s.indexOf(0); + return decoder.decode(end === -1 ? s : s.subarray(0, end)); +} + +function paxPath(data: Uint8Array): string | undefined { + // Records are " key=value\n". + for (const rec of decoder.decode(data).split('\n')) { + const m = /^\d+ path=(.*)$/.exec(rec); + if (m) return m[1]; + } + return undefined; +} + +/** Every regular file in a .tgz with its size, `package/` stripped. */ +export function tarEntries(tgz: Uint8Array): TarEntry[] { + const tar = gunzipSync(tgz); + const out: TarEntry[] = []; + let next: string | undefined; + for (let at = 0; at + 512 <= tar.length; ) { + const h = tar.subarray(at, at + 512); + if (h.every((x) => x === 0)) break; + const size = Number.parseInt(field(h, 124, 12).trim() || '0', 8); + const type = String.fromCharCode(h[156] ?? 0); + const data = tar.subarray(at + 512, at + 512 + size); + at += 512 + Math.ceil(size / 512) * 512; + if (type === 'L') next = field(data, 0, data.length); + else if (type === 'x') next = paxPath(data) ?? next; + else { + const prefix = field(h, 257, 6) === 'ustar' ? field(h, 345, 155) : ''; + const path = next ?? (prefix ? `${prefix}/${field(h, 0, 100)}` : field(h, 0, 100)); + next = undefined; + if (type === '0' || type === '\0') out.push({ path: path.replace(/^package\//, ''), size }); + } + } + return out; +} + +const mb = (n: number): string => `${(n / 1_000_000).toFixed(2)} MB`; + +function largest(entries: readonly TarEntry[]): string { + return [...entries] + .sort((a, b) => b.size - a.size) + .slice(0, 5) + .map((e) => `${e.path} (${mb(e.size)})`) + .join(', '); +} + +/** What breaks the budget (empty = fine); each message names the offenders and the fix. */ +export function budgetProblems( + entries: readonly TarEntry[], + packedBytes: number, + budget: PackBudget = PACK_BUDGET, +): string[] { + const problems: string[] = []; + const fix = + 'exclude what a consumer does not need in package.json "files", or raise PACK_BUDGET in scripts/pack-budget.ts with a reason'; + if (packedBytes > budget.packedBytes) + problems.push( + `tarball is ${mb(packedBytes)} packed, budget ${mb(budget.packedBytes)}; largest: ${largest(entries)}; ${fix}`, + ); + const unpacked = entries.reduce((s, e) => s + e.size, 0); + if (unpacked > budget.unpackedBytes) + problems.push( + `tarball is ${mb(unpacked)} unpacked, budget ${mb(budget.unpackedBytes)}; largest: ${largest(entries)}; ${fix}`, + ); + for (const e of entries) { + if (NEURAL.test(e.path)) + problems.push( + `ships a model output: ${e.path}; recipes/*/neural/ stays in git (keep "!recipes/*/neural" in package.json "files")`, + ); + else if (e.size > budget.fileBytes) + problems.push( + `ships a ${mb(e.size)} file (limit ${mb(budget.fileBytes)}): ${e.path}; exclude it in package.json "files"`, + ); + } + return problems; +} diff --git a/scripts/pack-smoke.ts b/scripts/pack-smoke.ts index a7460ba1..e9538ce2 100644 --- a/scripts/pack-smoke.ts +++ b/scripts/pack-smoke.ts @@ -2,8 +2,9 @@ // serves is what the gate tested: (1) `bunx t3d build` a two-box recipe to QA PASS; (2) import // every `exports` subpath under Bun and under Node (skipped with a notice when `node` is not on // PATH); (3) `tsc` a file importing every subpath with `skipLibCheck: false` and no DOM lib; -// (4) the tarball ships no tests, maps or demo PNGs, ships CLAUDE.md, NOTICE, every exports target and -// the files code loads by path at run time (the CLI bin, the path-tracer worker); (5) `t3d hero +// (4) the tarball ships no tests, maps, demo PNGs or recipes/*/neural/ model outputs, stays inside +// PACK_BUDGET (scripts/pack-budget.ts: packed, unpacked and per-file bytes), ships CLAUDE.md, NOTICE, +// every exports target and the files code loads by path at run time (the CLI bin, the path-tracer worker); (5) `t3d hero // --quality path` traces on worker threads from the installed package; (6) `t3d build --watch` // does its first rebuild through the installed CLI entry (then is killed); (7) a PNG and a contact // sheet hash the same under Bun and Node (with (2), skipped when `node` is not on PATH). @@ -17,6 +18,7 @@ import { tmpdir } from 'node:os'; import { join, resolve } from 'node:path'; import { isTs23dError, Ts23dError } from '../src/core/index.ts'; import { TIERS } from './lib/tiers.ts'; +import { budgetProblems, PACK_BUDGET, tarEntries } from './pack-budget.ts'; const NAME = '@developerz.ai/ts23d'; const ROOT = resolve(import.meta.dir, '..'); @@ -198,14 +200,19 @@ async function smoke(): Promise { const keys = Object.keys(pkg.exports); if (keys.join() !== expectedExports().join()) fail('4 tarball', `exports keys ${keys.join(', ')} ≠ ${expectedExports().join(', ')}`); - const listing = sh('4 tarball', ['tar', 'tzf', tarball], work) - .trim() - .split('\n') - .map((l) => l.replace(/^package\//, '')); - const problems = tarballProblems(listing, exportTargets(pkg.exports)); + const bytes = new Uint8Array(await Bun.file(tarball).arrayBuffer()); + const entries = tarEntries(bytes); + const problems = [ + ...tarballProblems( + entries.map((e) => e.path), + exportTargets(pkg.exports), + ), + ...budgetProblems(entries, bytes.length), + ]; if (problems.length > 0) fail('4 tarball', problems.join('\n')); + const unpacked = entries.reduce((s, e) => s + e.size, 0); console.log( - `✔ (4) tarball: ${listing.length} files, no tests/maps/demo PNGs, CLAUDE.md, NOTICE, bin + worker`, + `✔ (4) tarball: ${entries.length} files, ${(bytes.length / 1e6).toFixed(2)} MB packed / ${(unpacked / 1e6).toFixed(2)} MB unpacked (budget ${PACK_BUDGET.packedBytes / 1e6} / ${PACK_BUDGET.unpackedBytes / 1e6} MB), no neural outputs/tests/maps/demo PNGs, CLAUDE.md, NOTICE, bin + worker`, ); await Bun.write( diff --git a/scripts/recipe-costs-run.test.ts b/scripts/recipe-costs-run.test.ts new file mode 100644 index 00000000..83894486 --- /dev/null +++ b/scripts/recipe-costs-run.test.ts @@ -0,0 +1,89 @@ +import { describe, expect, it } from 'bun:test'; +import { costsFromRun, parseShardLog } from './recipe-costs-run.ts'; + +/** A gate job's log as GitHub serves it, cut down: shard `k` of 2 with the given lines. */ +const log = (k: number, verify: string, lines: readonly string[]): string => + [ + `$ bun test --timeout 120000 src scripts "--shard=${k}/2" "--timings=scripts/lib/test-timings.json"`, + `--shard=${k}/2: running 2/4 test files`, + ...lines, + `\u001b[32m✔ verify: ${verify}\u001b[0m`, + ] + .map((l, i) => `2026-11-19T09:00:${String(i).padStart(2, '0')}.0000000Z ${l}`) + .join('\n'); + +const LOG1 = log(1, 'lint 4000ms · test 3000ms · recipes 30000ms · pack 16000ms', [ + '##[group]src/a.test.ts:', + '(pass) a > one [400.00ms]', + '(pass) a > two [600.00ms]', + '##[endgroup]', + '##[group]src/b.test.ts:', + '(pass) b > one [1000.00ms]', + '##[endgroup]', + ' 3 pass', + 'Ran 3 tests across 2 files. [3.00s]', + '##[group]recipes/a/a.test.ts:', + '(pass) a > builds [5000.00ms]', + '##[endgroup]', + '✔ recipes (2 assets, 1 test files)', + ' → recipes/b/.out-tmp/b/sheet.png', + '✔ recipes/b 40.5s', +]); +const LOG2 = log(2, 'test 2000ms · recipes 10000ms', [ + '##[group]src/c.test.ts:', + '(pass) c > one [1000.00ms]', + '##[endgroup]', + '✔ recipes (1 assets, 0 test files)', + '✔ recipes/a 20s', +]); + +describe('costs from a CI run', () => { + const before = { + shard1: 1, + seconds: { 'gate:recipes/a': 30, 'gate:recipes/b': 9, 'recipes/_lib/x.test.ts': 10 }, + }; + const unitsOf = ({ k }: { k: number }): string[] => + k === 1 + ? ['gate:recipes/a', 'recipes/_lib/x.test.ts', 'shots:recipes/b'] + : ['gate:recipes/b', 'shots:recipes/a']; + const runs = [ + { ...parseShardLog(LOG1), job: 105.5 }, + { ...parseShardLog(LOG2), job: 44 }, + ]; + + it('refuses a log that is not a green gate job', () => { + expect(() => parseShardLog('##[group]Run bun run verify\n✗ verify failed')).toThrow(/fix:/); + }); + + it('reads the shard, the steps, the renders and the test files, not the recipe tests', () => { + expect(parseShardLog(LOG1)).toEqual({ + k: 1, + n: 2, + steps: { lint: 4000, test: 3000, recipes: 30000, pack: 16000 }, + shots: { 'recipes/b': 40.5 }, + tests: { 'src/a.test.ts': 1000, 'src/b.test.ts': 1000 }, + }); + }); + + it('refuses a missing shard and a run whose renders are not today’s split', () => { + expect(() => costsFromRun(runs.slice(0, 1), before, unitsOf)).toThrow(/each gate shard/); + const moved = (s: { k: number }): string[] => unitsOf({ k: 3 - s.k }); + expect(() => costsFromRun(runs, before, moved)).toThrow(/commit the run built/); + }); + + it('takes renders as logged, splits the recipes step by the old costs, scales tests to the step', () => { + expect(costsFromRun(runs, before, unitsOf)).toEqual({ + shard1: 20, // lint + pack + overhead: 12, // (105.5 − 53 − 40.5 + 44 − 12 − 20) / 2 + seconds: { + 'gate:recipes/a': 22.5, // 30 s of recipes step × 30 / (30 + 10) + 'gate:recipes/b': 10, + 'recipes/_lib/x.test.ts': 7.5, + 'shots:recipes/a': 20, + 'shots:recipes/b': 40.5, + }, + // 2 s of tests' own time in a 3 s step; one file in a 2 s step. + tests: { 'src/a.test.ts': 1500, 'src/b.test.ts': 1500, 'src/c.test.ts': 2000 }, + }); + }); +}); diff --git a/scripts/recipe-costs-run.ts b/scripts/recipe-costs-run.ts new file mode 100644 index 00000000..34f6bc3c --- /dev/null +++ b/scripts/recipe-costs-run.ts @@ -0,0 +1,127 @@ +// What one green `ci` run measured, as costs for scripts/lib/recipe-costs.json and +// scripts/lib/test-timings.json (`bun run scripts/recipe-costs.ts --from-run ` writes them). +// Per gate shard: +// measured a folder's renders (`✔ recipes/ 21.1s`, scripts/shots.ts); the shard's `test` +// and `recipes` steps and shard 1's own steps (`✔ verify: test 123ms · …`, +// scripts/verify.ts); the job's wall seconds (the jobs API) +// derived a folder's gate and a shared recipe test: the shard's `recipes` step seconds, split +// between its units in proportion to their previous costs (that step prints no time +// per folder, and its output arrives in one piece, so timestamps say nothing); +// a test file: the sum of its tests' `[ms]`, scaled to the shard's `test` step; +// `overhead`: the job's wall seconds minus its steps and renders (runner set-up, +// checkout, bun install, what changed, the stale check), averaged over the shards +// The split a run used is not in its log: it is recomputed from the committed costs, so ingest a +// run on the commit it built, before the costs change. A mismatch with the logged renders throws. + +import { type Costs, defaultCost, type Shard, shotsKey } from './lib/shard.ts'; +import { SHARDED } from './verify.ts'; + +/** What one gate job's log says: verify's steps (ms), renders (s per folder), tests (ms per file). */ +export interface ShardLog extends Shard { + readonly steps: Readonly>; + readonly shots: Readonly>; + readonly tests: Readonly>; +} + +/** A gate job: its log and its wall seconds. */ +export interface ShardRun extends ShardLog { + readonly job: number; +} + +/** A run's costs: seconds per unit, shard 1's own steps, the per-job overhead, ms per test file. */ +export interface RunCosts { + readonly seconds: Record; + readonly shard1: number; + readonly overhead: number; + readonly tests: Record; +} + +const sum = (xs: readonly number[]): number => xs.reduce((a, b) => a + b, 0); + +/** Read one gate job's log (a timestamp and colour codes may precede every line). */ +export function parseShardLog(log: string): ShardLog { + const shard = /--shard=(\d+)\/(\d+)/.exec(log); + const verify = /✔ verify: (.+)/.exec(log); + if (!shard || !verify) + throw new Error('no `--shard=k/n` and `✔ verify:` line. fix: pass a green gate job’s log'); + const steps: Record = {}; + for (const m of (verify[1] ?? '').matchAll(/(\S+) (\d+)ms/g)) steps[m[1] ?? ''] = Number(m[2]); + const shots: Record = {}; + for (const m of log.matchAll(/✔ (recipes\/\S+) ([\d.]+)s\s*$/gm)) + shots[m[1] ?? ''] = Number(m[2]); + const tests: Record = {}; + let file: string | undefined; + for (const line of log.split('\n')) { + const group = /##\[group\](\S+\.test\.tsx?):/.exec(line); + // The recipes step runs the tests under recipes/: they are part of that step's cost. + if (group) file = group[1]?.startsWith('recipes/') ? undefined : group[1]; + const ms = /\[([\d.]+)ms\]\s*$/.exec(line); + if (file && (group || ms)) tests[file] = (tests[file] ?? 0) + Number(ms?.[1] ?? 0); + } + return { k: Number(shard[1]), n: Number(shard[2]), steps, shots, tests }; +} + +/** + * The costs a run measured (the header says which are measured and which derived). `unitsOf` is + * the split the run used: today's, so each shard's logged renders must be the ones it gives it. + */ +export function costsFromRun( + runs: readonly ShardRun[], + old: Costs, + unitsOf: (shard: Shard) => readonly string[], +): RunCosts { + const n = runs[0]?.n ?? 0; + if (n === 0 || runs.length !== n || new Set(runs.map((r) => (r.n === n ? r.k : 0))).size !== n) + throw new Error(`need each gate shard's log once, got ${runs.length}. fix: pass a green run`); + const seconds: Record = {}; + const tests: Record = {}; + let shard1 = 0; + let overhead = 0; + for (const r of runs) { + const units = unitsOf(r); + const gated = units.filter((u) => !u.startsWith(shotsKey(''))); + const rendered = units.filter((u) => u.startsWith(shotsKey(''))).sort(); + const recipes = r.steps['recipes']; + const logged = Object.keys(r.shots).map(shotsKey).sort(); + if (recipes === undefined || logged.join() !== rendered.join()) + throw new Error( + `shard ${r.k}/${n} did not gate and render the units today's costs give it. fix: check out the commit the run built; a docs-only run measures nothing`, + ); + for (const [dir, s] of Object.entries(r.shots)) seconds[shotsKey(dir)] = s; + // A unit with no cost yet (or 0) still gets a share. + const weight = (u: string): number => Math.max(old.seconds[u] ?? defaultCost(old), 0.1); + const weights = sum(gated.map(weight)); + for (const u of gated) seconds[u] = (recipes / 1000) * (weight(u) / weights); + const testMs = sum(Object.values(r.tests)); + for (const [f, ms] of Object.entries(r.tests)) + tests[f] = Math.round(testMs > 0 ? (ms / testMs) * (r.steps['test'] ?? testMs) : 0); + for (const [step, ms] of Object.entries(r.steps)) + if (r.k === 1 && !SHARDED.some((s) => s === step)) shard1 += ms / 1000; + const work = sum(Object.values(r.steps)) / 1000 + sum(Object.values(r.shots)); + overhead += (r.job - work) / n; + } + return { seconds, shard1, overhead, tests }; +} + +/** `gh api` on this repository; throws with gh's own message when it fails. */ +function gh(args: readonly string[]): string { + const p = Bun.spawnSync(['gh', 'api', '--allow-escape-sequences', ...args], { stderr: 'pipe' }); + if (p.exitCode !== 0) throw new Error(`gh api ${args.join(' ')}: ${p.stderr} fix: gh auth login`); + return p.stdout.toString(); +} + +/** Every gate job of a workflow run: its parsed log and its wall seconds. */ +export function fetchRun(id: string): ShardRun[] { + const jq = + '.jobs[] | select(.name | startswith("gate")) | "\\(.id) \\(.started_at) \\(.completed_at)"'; + const rows = gh(['--paginate', `repos/{owner}/{repo}/actions/runs/${id}/jobs`, '--jq', jq]); + return rows + .trim() + .split('\n') + .filter((row) => row !== '') + .map((row) => { + const [job = '', start = '', end = ''] = row.split(' '); + const log = gh([`repos/{owner}/{repo}/actions/jobs/${job}/logs`]); + return { ...parseShardLog(log), job: (Date.parse(end) - Date.parse(start)) / 1000 }; + }); +} diff --git a/scripts/recipe-costs.test.ts b/scripts/recipe-costs.test.ts index 43de3488..bc99b7b2 100644 --- a/scripts/recipe-costs.test.ts +++ b/scripts/recipe-costs.test.ts @@ -1,5 +1,5 @@ import { describe, expect, it } from 'bun:test'; -import { mergeCosts, plan, SHARD1_STEPS } from './recipe-costs.ts'; +import { mergeCosts, pickShards, plan, SHARD1_STEPS, worstJob } from './recipe-costs.ts'; describe('recipe costs', () => { const old = { shard1: 40, seconds: { 'recipes/a': 10, 'recipes/gone': 5 } }; @@ -11,14 +11,20 @@ describe('recipe costs', () => { expect(partial).toEqual({ shard1: 40, seconds: { 'recipes/a': 12, 'recipes/gone': 5 } }); }); - it('writes keys sorted so a refresh diffs cleanly', () => { - const m = mergeCosts({ shard1: 0, seconds: {} }, { 'recipes/z': 1, 'recipes/a': 2 }, 0); + it('writes keys sorted so a refresh diffs cleanly, and keeps the overhead it did not measure', () => { + const m = mergeCosts( + { shard1: 0, overhead: 16, seconds: {} }, + { 'recipes/z': 1, 'recipes/a': 2 }, + 0, + ); expect(Object.keys(m.seconds)).toEqual(['recipes/a', 'recipes/z']); + expect(m.overhead).toBe(16); }); - it('predicts a shard as its units plus an even share of the tests', () => { + it('predicts a job as the overhead, its units and an even share of the tests', () => { const costs = { shard1: 10, seconds: { a: 30, b: 20 } }; expect(plan(costs, ['a', 'b'], 40, 2)).toEqual([10 + 20 + 20, 30 + 20]); + expect(plan({ ...costs, overhead: 5 }, ['a', 'b'], 40, 2)).toEqual([55, 55]); }); it('times only the steps shard 1 runs alone', () => { @@ -28,3 +34,19 @@ describe('recipe costs', () => { expect(SHARD1_STEPS).toContain('pack'); }); }); + +describe('shard count', () => { + const costs = { shard1: 0, overhead: 10, seconds: { a: 100, b: 100, c: 100, d: 100 } }; + const units = ['a', 'b', 'c', 'd']; + + it('counts the longest test file against the heaviest shard', () => { + expect(worstJob([110, 110], [30, 10], 2)).toBe(110 - 20 + 30); + }); + + it('picks the fewest shards whose worst job fits, or none', () => { + expect(pickShards(costs, units, [20, 20], 240)).toBe(2); // 10 + 200 + 20 + // Three shards do not help: one of them still holds two units. + expect(pickShards(costs, units, [20, 20], 229)).toBe(4); + expect(pickShards(costs, units, [20, 20], 100)).toBeUndefined(); // one unit + overhead = 110 + }); +}); diff --git a/scripts/recipe-costs.ts b/scripts/recipe-costs.ts index 6e38c27d..450ae334 100644 --- a/scripts/recipe-costs.ts +++ b/scripts/recipe-costs.ts @@ -1,28 +1,56 @@ -// Measure what each unit of sharded CI work costs and write scripts/lib/recipe-costs.json, which -// scripts/lib/shard.ts balances T3D_SHARD by. Sequential and wall-clock, so run it on a quiet box. -// bun run scripts/recipe-costs.ts every recipe folder's gate and, separately, its -// renders (`bun run shots`, which rewrites out/), -// every shared recipes/_*/ test and shard 1's checks -// bun run scripts/recipe-costs.ts t34 wolf only these folders, merged into the file -// bun run scripts/recipe-costs.ts --tests also refresh scripts/lib/test-timings.json, which -// balances the sharded `test` step (bun test --shard) -// bun run scripts/recipe-costs.ts --plan measure nothing: print the predicted seconds per -// shard for 4..20 shards from the committed files -// Seconds are this machine's; only their ratios matter for balance. Exit 1 when a measured command -// failed (its time is still recorded: a red recipe costs what it costs). +// Record what each unit of sharded CI work costs, in GitHub seconds, in scripts/lib/recipe-costs.json +// (scripts/lib/shard.ts balances T3D_SHARD by it) and scripts/lib/test-timings.json (`bun test +// --shard` balances by it). +// bun run scripts/recipe-costs.ts --from-run the calibration: read a green `ci` run's gate +// logs (needs `gh`) and rewrite both files from what +// GitHub measured. Run it on the commit the run built. +// bun run scripts/recipe-costs.ts --plan measure nothing: predicted seconds per job for +// 4..20 shards; → is CI_SHARDS, ★ the count to pick +// bun run scripts/recipe-costs.ts t34 wolf no run yet (new folders): time these folders' gate +// and renders here and merge an estimate into the file +// bun run scripts/recipe-costs.ts --renders time only every folder's renders here, merged +// bun run scripts/recipe-costs.ts time everything here (sequential: use a quiet box) +// bun run scripts/recipe-costs.ts --tests also re-time every test file here +// `--from-run` is scripts/recipe-costs-run.ts, which says what it measures and what it derives. +// Times taken here are estimates of GitHub seconds: renders times CI_RENDER_FACTOR, gates as they +// are. Exit 1 when a measured command failed (its time is still recorded). import { recipeFolders, tsFiles } from './lib/files.ts'; -import { CI_SHARDS, COSTS, type Costs, gateKey, shardLoads, shotsKey } from './lib/shard.ts'; +import { + CI_SHARDS, + COSTS, + type Costs, + gateKey, + inShard, + type Shard, + shardLoads, + shotsKey, +} from './lib/shard.ts'; +import { costsFromRun, fetchRun, type RunCosts } from './recipe-costs-run.ts'; import { sharedRecipeTests } from './recipes.ts'; import { SHARDED, STEPS, TEST_TIMINGS as TEST_TIMINGS_FILE } from './verify.ts'; export const COSTS_FILE = 'scripts/lib/recipe-costs.json'; +/** + * A folder's renders on a 4-vCPU GitHub runner over the same renders on the 16-thread dev machine: + * the median over the 55 folders of run 44492321426 (2026-11-19; they range 0.9 to 2.3, which is + * why `--from-run` is the calibration). Gates ran at about this machine's speed there (0.95×). + */ +export const CI_RENDER_FACTOR = 1.5; + +/** The slowest job `--plan` accepts (seconds, overhead included): push → green stays under 5 min. */ +export const JOB_LIMIT = 240; +/** With the `site` job this is the 20 jobs a free account runs at once. */ +export const MAX_SHARDS = 19; + /** The steps only shard 1 runs: every gate step except the sharded `test` and `recipes`. */ export const SHARD1_STEPS: readonly string[] = STEPS.map((s) => s.name).filter( (n) => !SHARDED.includes(n), ); +const round = (s: number): number => Math.round(s * 10) / 10; + /** * New measurements over old ones. A full run (`all` = every current unit) also drops units that * no longer exist; a partial run keeps everything it did not measure. Keys sorted, 0.1 s. @@ -32,36 +60,92 @@ export function mergeCosts( measured: Readonly>, shard1: number | undefined, all?: readonly string[], + overhead: number | undefined = old.overhead, ): Costs { const merged: Record = { ...old.seconds, ...measured }; const keep = all ? new Set(all) : undefined; const seconds: Record = {}; for (const k of Object.keys(merged).sort()) - if (!keep || keep.has(k)) seconds[k] = Math.round((merged[k] ?? 0) * 10) / 10; - return { shard1: Math.round((shard1 ?? old.shard1) * 10) / 10, seconds }; + if (!keep || keep.has(k)) seconds[k] = round(merged[k] ?? 0); + return { + shard1: round(shard1 ?? old.shard1), + ...(overhead === undefined ? {} : { overhead: round(overhead) }), + seconds, + }; } -/** Predicted seconds per shard: its recipe units (+ shard 1's checks) and 1/n of the tests. */ +/** Predicted seconds per job: overhead, its recipe units (+ shard 1's checks) and 1/n of the tests. */ export function plan( costs: Costs, units: readonly string[], testSeconds: number, n: number, ): number[] { - return shardLoads(units, n, costs).map((s) => s + testSeconds / n); + return shardLoads(units, n, costs).map((s) => s + (costs.overhead ?? 0) + testSeconds / n); +} + +/** + * The slowest job when the heaviest recipe shard also draws the heaviest test slice: `bun test + * --shard` splits on its own, and no slice is shorter than the longest test file. + */ +export function worstJob(loads: readonly number[], tests: readonly number[], n: number): number { + const total = tests.reduce((a, b) => a + b, 0); + return Math.max(...loads) - total / n + Math.max(total / n, ...tests); +} + +/** The fewest shards (fewest runner-minutes: every job pays the overhead) whose worst job fits. */ +export function pickShards( + costs: Costs, + units: readonly string[], + tests: readonly number[], + limit = JOB_LIMIT, +): number | undefined { + const total = tests.reduce((a, b) => a + b, 0); + for (let n = 1; n <= MAX_SHARDS; n++) + if (worstJob(plan(costs, units, total, n), tests, n) <= limit) return n; + return undefined; } async function printPlan(units: readonly string[]): Promise { const t = (await Bun.file(TEST_TIMINGS_FILE).json()) as { files: Record }; - const tests = Object.values(t.files).reduce((a, b) => a + b, 0) / 1000; + const tests = Object.values(t.files).map((ms) => ms / 1000); + const total = tests.reduce((a, b) => a + b, 0); + const pick = pickShards(COSTS, units, tests); for (let n = 4; n <= 20; n++) { - const loads = plan(COSTS, units, tests, n); - const max = Math.max(...loads); + const loads = plan(COSTS, units, total, n); const sum = loads.reduce((a, b) => a + b, 0); + const mark = `${n === CI_SHARDS ? '→' : ' '}${n === pick ? '★' : ' '}`; console.log( - `${n === CI_SHARDS ? '→' : ' '} ${String(n).padStart(2)} shards: max ${max.toFixed(0)}s · mean ${(sum / n).toFixed(0)}s · total ${sum.toFixed(0)}s · ${loads.map((l) => l.toFixed(0)).join(' ')}`, + `${mark} ${String(n).padStart(2)} shards: max ${Math.max(...loads).toFixed(0)}s · worst ${worstJob(loads, tests, n).toFixed(0)}s · mean ${(sum / n).toFixed(0)}s · ${(sum / 60).toFixed(1)} runner-min · ${loads.map((l) => l.toFixed(0)).join(' ')}`, ); } + const [top] = units.map((u) => [COSTS.seconds[u] ?? 0, u] as const).sort((a, b) => b[0] - a[0]); + console.log( + `seconds per job, ${COSTS.overhead ?? 0}s overhead included; worst = the heaviest shard with the longest test slice (${Math.max(...tests).toFixed(0)}s); ★ = fewest shards ≤ ${MAX_SHARDS} with worst ≤ ${JOB_LIMIT}s${pick ? '' : ' (none)'}; largest unit: ${top?.[1]} ${top?.[0]}s`, + ); +} + +async function fromRun(id: string, unitsOf: (shard: Shard) => readonly string[]): Promise { + if (!/^\d+$/.test(id)) { + console.error('✗ --from-run needs a run id. fix: gh run list --workflow ci.yml -s success'); + return 2; + } + let run: RunCosts; + try { + run = costsFromRun(fetchRun(id), COSTS, unitsOf); + } catch (e) { + console.error(`✗ run ${id}: ${e instanceof Error ? e.message : String(e)}`); + return 1; + } + const costs = mergeCosts(COSTS, run.seconds, run.shard1, Object.keys(run.seconds), run.overhead); + await Bun.write(COSTS_FILE, `${JSON.stringify(costs, null, 2)}\n`); + const files = Object.fromEntries( + Object.entries(run.tests).sort((a, b) => b[1] - a[1] || (a[0] < b[0] ? -1 : 1)), + ); + await Bun.write(TEST_TIMINGS_FILE, `${JSON.stringify({ version: 1, files }, null, 2)}\n`); + console.log(`→ ${COSTS_FILE} (shard1 ${costs.shard1}s, overhead ${costs.overhead}s)`); + console.log(`→ ${TEST_TIMINGS_FILE} (${Object.keys(files).length} files)`); + return 0; } /** Run a command with T3D_SHARD unset; its wall seconds and whether it exited 0. */ @@ -77,7 +161,8 @@ function timed(cmd: readonly string[]): { s: number; ok: boolean } { async function main(): Promise { const args = process.argv.slice(2); - const names = args.filter((a) => !a.startsWith('--')); + const run = args.indexOf('--from-run'); + const names = args.filter((a, i) => !a.startsWith('--') && (run < 0 || i !== run + 1)); const { folders } = await recipeFolders(); const shared = sharedRecipeTests(await tsFiles('recipes')); const unknown = names.filter((n) => !folders.some((f) => f.dir === `recipes/${n}`)); @@ -90,6 +175,7 @@ async function main(): Promise { ...folders.map((f) => shotsKey(f.dir)), ...shared, ]; + if (run >= 0) return fromRun(args[run + 1] ?? '', (s) => inShard(units, s, (u) => u)); if (args.includes('--plan')) { await printPlan(units); return 0; @@ -97,18 +183,23 @@ async function main(): Promise { const picked = names.length ? folders.filter((f) => names.some((n) => f.dir === `recipes/${n}`)) : folders; + const rendersOnly = args.includes('--renders'); const measured: Record = {}; let failed = 0; for (const { dir } of picked) { const name = dir.slice('recipes/'.length); - const gate = timed(['bun', 'run', 'scripts/recipes.ts', name]); + if (!rendersOnly) { + const gate = timed(['bun', 'run', 'scripts/recipes.ts', name]); + measured[gateKey(dir)] = gate.s; + failed += Number(!gate.ok); + } const shots = timed(['bun', 'run', 'scripts/shots.ts', name]); - measured[gateKey(dir)] = gate.s; - measured[shotsKey(dir)] = shots.s; - failed += Number(!gate.ok) + Number(!shots.ok); + measured[shotsKey(dir)] = shots.s * CI_RENDER_FACTOR; + failed += Number(!shots.ok); } + const partial = names.length > 0 || rendersOnly; let shard1: number | undefined; - if (!names.length) { + if (!partial) { for (const t of shared) { const r = timed(['bun', 'test', '--timeout', '120000', `./${t}`]); measured[t] = r.s; @@ -118,17 +209,13 @@ async function main(): Promise { shard1 = r.s; failed += Number(!r.ok); } - const file = Bun.file(COSTS_FILE); - const old: Costs = (await file.exists()) ? await file.json() : { shard1: 0, seconds: {} }; - const all = names.length ? undefined : units; + const all = partial ? undefined : units; await Bun.write( COSTS_FILE, - `${JSON.stringify(mergeCosts(old, measured, shard1, all), null, 2)}\n`, + `${JSON.stringify(mergeCosts(COSTS, measured, shard1, all), null, 2)}\n`, ); console.log(`→ ${COSTS_FILE}`); if (args.includes('--tests')) { - if (!(await Bun.file(TEST_TIMINGS_FILE).exists())) - await Bun.write(TEST_TIMINGS_FILE, '{ "version": 1, "files": {} }\n'); const r = timed(['bun', 'run', 'test', `--timings=${TEST_TIMINGS_FILE}`, '--update-timings']); failed += Number(!r.ok); console.log(`→ ${TEST_TIMINGS_FILE}`); diff --git a/scripts/shots-folder.ts b/scripts/shots-folder.ts index df619c21..b4fca689 100644 --- a/scripts/shots-folder.ts +++ b/scripts/shots-folder.ts @@ -1,14 +1,14 @@ // One recipe folder's committed renders in ONE process: `t3d shot`, `hero` (path traced), `blueprint` // and `icon` run in-process through the CLI's own `run()`, so the recipe and the library load once -// instead of four times. The output bytes are the four separate commands' bytes (same code path, -// same flags; on 2026-11-19 all 55 folders matched the four-process output). It saves only the -// imports (human_head 12 s each, most recipes about 1 s): each command still compiles the asset -// itself (human_head LOD0 about 55 s per command). Sharing one compile across the four commands -// needs a src/cli change: a per-process compile cache or one multi-output command. +// instead of four times, and the four commands share one compile per asset and LOD +// (`shareCompiles`, src/cli/compile-cache.ts): before that each command compiled LOD0 again +// (human_head about 55 s per command). The output bytes are the four separate commands' bytes +// (same code path, same flags; the ci `gate` job fails on any difference from the committed out/). // One process per folder, never one for every folder: recipes register materials globally and the // painted trim sheet follows the registered set, so another recipe's materials would change pixels. // bun run scripts/shots-folder.ts recipes/t34/t34.ts recipes/t34/.out-tmp (scripts/shots.ts runs it) +import { shareCompiles } from '../src/cli/compile-cache.ts'; import { run } from '../src/cli/run.ts'; import { HERO_FLAGS } from './shots.ts'; @@ -27,13 +27,21 @@ export const renderArgv = (file: string, out: string): string[][] => /** Runs one `t3d` argv; resolves its exit code. */ export type Cli = (argv: readonly string[]) => Promise; -/** Every render in order; stops at the first non-zero exit and returns it (0 when all passed). */ +/** + * Every render in order, sharing compiles between them; stops at the first non-zero exit and + * returns it (0 when all passed). + */ export async function renderFolder(file: string, out: string, cli: Cli = run): Promise { - for (const argv of renderArgv(file, out)) { - const code = await cli(argv); - if (code !== 0) return code; + const stop = shareCompiles(); + try { + for (const argv of renderArgv(file, out)) { + const code = await cli(argv); + if (code !== 0) return code; + } + return 0; + } finally { + stop(); } - return 0; } if (import.meta.main) { diff --git a/scripts/shots.test.ts b/scripts/shots.test.ts index 0656c398..b84d8fb1 100644 --- a/scripts/shots.test.ts +++ b/scripts/shots.test.ts @@ -2,6 +2,8 @@ import { afterAll, describe, expect, it } from 'bun:test'; import { existsSync } from 'node:fs'; import { mkdir, rm } from 'node:fs/promises'; import { join } from 'node:path'; +import { compileOnce } from '../src/cli/compile-cache.ts'; +import { box, defineAsset } from '../src/index.ts'; import { HERO_FLAGS, type Runner, refresh } from './shots.ts'; import { type Cli, renderArgv, renderFolder } from './shots-folder.ts'; @@ -68,4 +70,27 @@ describe('shots in one process per folder', () => { expect(seen).toEqual(['shot', 'hero']); expect(await renderFolder('x.ts', 'o', async () => 0)).toBe(0); }); + + it('shares one compile between the renders, and none once the folder is done', async () => { + let builds = 0; + const asset = defineAsset({ + name: 'shared_by_shots', + kind: 'prop', + budget: { triangles: [100, 100, 100], draws: 1 }, + build(ctx) { + builds++; + ctx.part('b', box({ size: [1, 1, 1], color: [1, 1, 1] })); + }, + }); + const cli: Cli = async ([cmd]) => { + compileOnce(asset, { lod: 0 }); + return cmd === 'blueprint' ? 1 : 0; + }; + expect(await renderFolder('x.ts', 'o', cli)).toBe(1); + expect(builds).toBe(1); + // Off again, also after a failed render: every call compiles. + compileOnce(asset, { lod: 0 }); + compileOnce(asset, { lod: 0 }); + expect(builds).toBe(3); + }); }); diff --git a/scripts/shots.ts b/scripts/shots.ts index 562bd420..d5f2f14b 100644 --- a/scripts/shots.ts +++ b/scripts/shots.ts @@ -5,7 +5,8 @@ // bun run shots t34 tiger_1 only these folders // Each folder renders into a sibling temp dir first and replaces out/ only when every step // succeeded: a failure (a half-edited shared module) keeps the old outputs. One `bun` process per -// folder (scripts/shots-folder.ts) runs all four renders from a single import of the recipe. +// folder (scripts/shots-folder.ts) runs all four renders from a single import of the recipe and +// one compile per asset and LOD. // T3D_SHARD=k/n renders only this shard's folders (scripts/lib/shard.ts, balanced by measured cost). import { readdir, rename, rm } from 'node:fs/promises'; @@ -19,10 +20,10 @@ import { sharedRecipeTests } from './recipes.ts'; /** * The committed hero is what critiques judge, so it is path traced (src/render/path/): 960 × 540 * at 16 samples per pixel + denoise. Measured on 4 threads: wolf (138k tris) 6.7 s of tracing, - * human_head 6 s, roadster_v10 (900k tris) 13 s; a 4-vCPU CI runner ≈ 1.5× that. Tracing is not - * what makes the shots job slow: each of the four renders compiles the asset again (CI run - * 44460876935: human_head 436 s for its four renders; face_demo's five LOD0 compiles take about - * 50 s each time). Raise spp only with a fresh measurement. + * human_head 6 s, roadster_v10 (900k tris) 13 s; a 4-vCPU CI runner ≈ 1.5× that. Tracing was never + * what made the shots job slow: each of the four renders used to compile the asset again (CI run + * 44460876935: human_head 436 s for its four renders); they now share one compile + * (src/cli/compile-cache.ts). Raise spp only with a fresh measurement. */ export const HERO_FLAGS: readonly string[] = [ '--quality', diff --git a/src/cli/build-one.ts b/src/cli/build-one.ts index 4886d5d6..9311dc84 100644 --- a/src/cli/build-one.ts +++ b/src/cli/build-one.ts @@ -6,7 +6,7 @@ import { mkdir } from 'node:fs/promises'; import { join } from 'node:path'; -import { type Asset, type CompiledAsset, compile, type Lod } from '../core/index.ts'; +import type { Asset, CompiledAsset, Lod } from '../core/index.ts'; import { drawFindings, exportLodGlb, @@ -18,6 +18,7 @@ import type { PaintedSheet } from '../material/index.ts'; import { codeOf, type Finding, runLodQa, runQa } from '../qa/index.ts'; import { beauty, contactSheet, thumbnail } from '../render/index.ts'; import type { Args } from './args.ts'; +import { compileOnce } from './compile-cache.ts'; import { type AssetManifest, manifestOf } from './manifest.ts'; import { variantOf } from './render-cmds.ts'; import { foldLods } from './report.ts'; @@ -39,7 +40,7 @@ export async function buildOne( ): Promise { const dir = join(args.out, asset.name); await mkdir(dir, { recursive: true }); - const lods = args.lods.map((lod: Lod) => compile(asset, { lod })); + const lods = args.lods.map((lod: Lod) => compileOnce(asset, { lod })); const findings = qaOf(asset, lods); // Manifest paths are relative to `dir`; `write` returns the relative name. const write = async (name: string, bytes: Uint8Array | string): Promise => { @@ -86,7 +87,7 @@ export async function buildOne( // Images show the chosen livery; the GLBs above carry every variant regardless. const variant = variantOf(asset, args); const look = (lod: Lod, compiled?: CompiledAsset): CompiledAsset => - variant ? compile(asset, { lod, variant }) : (compiled ?? compile(asset, { lod })); + variant ? compileOnce(asset, { lod, variant }) : (compiled ?? compileOnce(asset, { lod })); const png = (name: string): string => (variant ? `${name}_${variant}.png` : `${name}.png`); let hero: string | undefined; let thumb: string | undefined; diff --git a/src/cli/compile-cache.test.ts b/src/cli/compile-cache.test.ts new file mode 100644 index 00000000..524c0926 --- /dev/null +++ b/src/cli/compile-cache.test.ts @@ -0,0 +1,189 @@ +import { afterEach, expect, it } from 'bun:test'; +import { mkdtemp, rm } from 'node:fs/promises'; +import { tmpdir } from 'node:os'; +import { join } from 'node:path'; +import { + box, + type CompiledAsset, + compile, + cylinder, + defineAsset, + paintTrimSheet, +} from '../index.ts'; +import { DEFAULT_SHEET } from '../material/index.ts'; +import { parseArgs } from './args.ts'; +import { buildOne } from './build-one.ts'; +import { compileKey, compileOnce, shareCompiles, sharedCompiles } from './compile-cache.ts'; +import { iconOne } from './icon-cmd.ts'; +import { blueprintOne, heroOne } from './render-cmds.ts'; + +let builds = 0; +const drum = (name: string) => + defineAsset({ + name, + kind: 'prop', + seed: 7, + budget: { triangles: [2000, 2000, 2000], draws: 1 }, + paint: { body: '#ffffff' }, + variants: { blue: { body: '#2222aa' } }, + build(ctx) { + builds++; + ctx.part( + 'drum', + cylinder({ radius: 0.3, length: 1, segments: ctx.pick([32, 16, 8]), color: [1, 1, 1] }), + { paint: 'body' }, + ); + }, + }); + +let stop: (() => void) | undefined; +afterEach(() => { + stop?.(); + stop = undefined; +}); + +/** The vertex array two views of one compile share. */ +const verts = (a: CompiledAsset): Float32Array | undefined => a.parts[0]?.mesh?.positions; + +/** Every byte a consumer could change: typed arrays as hex, the rest as JSON. */ +const fingerprint = (a: CompiledAsset): string => + new Bun.CryptoHasher('sha256') + .update( + JSON.stringify(a, (_k, v: unknown) => + ArrayBuffer.isView(v) + ? Buffer.from(v.buffer, v.byteOffset, v.byteLength).toString('hex') + : typeof v === 'function' + ? String(v) + : v, + ), + ) + .digest('hex'); + +it('is plain compile until shareCompiles(): every call compiles again', () => { + const a = drum('off'); + builds = 0; + expect(verts(compileOnce(a, { lod: 0 }))).not.toBe(verts(compileOnce(a, { lod: 0 }))); + expect(builds).toBe(2); + expect(sharedCompiles()).toEqual({ count: 0, bytes: 0 }); +}); + +it('misses on another LOD, seed or variant, and hits on the same ones', () => { + stop = shareCompiles(); + const a = drum('keys'); + builds = 0; + const lod0 = compileOnce(a, { lod: 0 }); + expect(verts(compileOnce(a, { lod: 0 }))).toBe(verts(lod0)); + expect(verts(compileOnce(a))).toBe(verts(lod0)); // the default LOD is 0 + expect(verts(compileOnce(a, { seed: 7 }))).toBe(verts(lod0)); // the asset's own seed + expect(builds).toBe(1); + expect(verts(compileOnce(a, { lod: 2 }))).not.toBe(verts(lod0)); + expect(verts(compileOnce(a, { lod: 0, seed: 8 }))).not.toBe(verts(lod0)); + const blue = compileOnce(a, { lod: 0, variant: 'blue' }); + expect(blue.variant).toBe('blue'); + expect(verts(blue)).not.toBe(verts(lod0)); + expect(verts(compileOnce(a, { lod: 0, variant: 'blue' }))).toBe(verts(blue)); + expect(builds).toBe(4); + expect(compileKey(a, { lod: 0 })).not.toBe(compileKey(a, { lod: 0, variant: 'blue' })); +}); + +it('never hands one asset another asset’s compile, even under the same name', () => { + stop = shareCompiles(); + const first = drum('twin'); + const again = drum('twin'); // what a re-imported (edited) recipe is: a new object + const a = compileOnce(first, { lod: 0 }); + const b = compileOnce(again, { lod: 0 }); + expect(verts(b)).not.toBe(verts(a)); + expect(verts(compileOnce(first, { lod: 0 }))).toBe(verts(a)); + expect(verts(compileOnce(again, { lod: 0 }))).toBe(verts(b)); +}); + +it('hands every caller its own asset, parts and mesh objects over the shared vertex arrays', () => { + stop = shareCompiles(); + const a = drum('views'); + const one = compileOnce(a, { lod: 0 }); + const two = compileOnce(a, { lod: 0 }); + // What QA and the renderers memoise on (world meshes per parts array, grids per mesh object). + expect(two).not.toBe(one); + expect(two.parts).not.toBe(one.parts); + expect(two.parts[0]).not.toBe(one.parts[0]); + expect(two.parts[0]?.mesh).not.toBe(one.parts[0]?.mesh); + expect(verts(two)).toBe(verts(one)); + expect(two).toEqual(one); +}); + +it('never shares a compile made with tuned trial values', () => { + stop = shareCompiles(); + const a = drum('tuned'); + builds = 0; + compileOnce(a, { lod: 0, tuned: {} }); + compileOnce(a, { lod: 0, tuned: {} }); + expect(builds).toBe(2); + expect(sharedCompiles().count).toBe(0); +}); + +it('does not keep a compile that threw', () => { + stop = shareCompiles(); + let fail = true; + const a = defineAsset({ + name: 'flaky', + kind: 'prop', + budget: { triangles: [100, 100, 100], draws: 1 }, + build(ctx) { + if (fail) throw new RangeError('first time'); + ctx.part('b', box({ size: [1, 1, 1], color: [1, 1, 1] })); + }, + }); + expect(() => compileOnce(a)).toThrow('first time'); + fail = false; + expect(compileOnce(a).parts.length).toBe(1); +}); + +it('drops the oldest compiles over the byte limit and forgets everything on stop', () => { + const a = drum('bounded'); + const one = compile(a, { lod: 0 }); + const bytes = Object.values(one.parts[0]?.mesh ?? {}).reduce( + (n, v) => n + (ArrayBuffer.isView(v) ? v.byteLength : 0), + 0, + ); + stop = shareCompiles(bytes * 2); // room for two LOD0-sized compiles + const first = compileOnce(a, { lod: 0 }); + compileOnce(a, { lod: 2 }); + expect(sharedCompiles()).toEqual({ count: 2, bytes: expect.any(Number) }); + expect(sharedCompiles().bytes).toBeLessThanOrEqual(bytes * 2); + compileOnce(a, { lod: 0, variant: 'blue' }); // pushes the first LOD0 out + expect(sharedCompiles().bytes).toBeLessThanOrEqual(bytes * 2); + expect(verts(compileOnce(a, { lod: 0 }))).not.toBe(verts(first)); + stop(); + expect(sharedCompiles()).toEqual({ count: 0, bytes: 0 }); + expect(verts(compileOnce(a, { lod: 2 }))).not.toBe(verts(compileOnce(a, { lod: 2 }))); +}); + +it('shot, hero, blueprint and icon share one compile per LOD and leave it untouched', async () => { + const out = await mkdtemp(join(tmpdir(), 't3d-compile-cache-')); + try { + stop = shareCompiles(); + const a = drum('shared'); + const sheet = paintTrimSheet({ ...DEFAULT_SHEET, size: 512 }); + const lod0 = compileOnce(a, { lod: 0 }); + const lod2 = compileOnce(a, { lod: 2 }); + const before = [fingerprint(lod0), fingerprint(lod2)]; + builds = 0; + const size = ['--size', '96x54']; + await buildOne(a, parseArgs(['shot', 'x.ts', '--out', out]), undefined, undefined); + await heroOne(a, parseArgs(['hero', 'x.ts', '--out', out, ...size]), sheet); + await heroOne( + a, + parseArgs(['hero', 'x.ts', '--out', out, ...size, '--quality', 'path', '--spp', '1']), + sheet, + ); + await blueprintOne(a, parseArgs(['blueprint', 'x.ts', '--out', out])); + await iconOne(a, parseArgs(['icon', 'x.ts', '--out', out]), sheet); + expect(builds).toBe(1); // only shot's LOD1; LOD0 and LOD2 were already held + expect(verts(compileOnce(a, { lod: 0 }))).toBe(verts(lod0)); + expect([fingerprint(lod0), fingerprint(lod2)]).toEqual(before); + // The same bytes as a compile nobody has used. + expect(fingerprint(lod0)).toBe(fingerprint(compile(a, { lod: 0 }))); + } finally { + await rm(out, { recursive: true, force: true }); + } +}); diff --git a/src/cli/compile-cache.ts b/src/cli/compile-cache.ts new file mode 100644 index 00000000..65fc3fc3 --- /dev/null +++ b/src/cli/compile-cache.ts @@ -0,0 +1,104 @@ +// One compile per (asset, LOD, seed, variant) for a process that runs several `t3d` commands on the +// same recipe: `bun run shots` runs shot, hero, blueprint and icon through `run()` in one process, +// and each of them used to compile LOD0 again (human_head: about 55 s each time). +// Off by default: `compileOnce` is plain `compile` until `shareCompiles()` is called, so `t3d mcp` +// (one long-lived process, where the recipe's committed neural outputs, the import cache and +// tune.json can change between two calls) never sees an old compile. A caller that turns it on +// promises those inputs stay as they are until it calls the returned `stop`. +// Keyed on the asset OBJECT (a re-imported recipe is a new object, so a miss), the LOD, the seed +// in effect and the variant; a compile with `tuned` trial values is never shared. Consumers treat +// a CompiledAsset as read-only (compile-cache.test.ts fingerprints one around every consumer). +// Memory: every caller gets its own VIEW (new asset, parts and mesh objects over the same vertex +// arrays). QA and the renderers memoise world meshes, triangle grids, shells and solids per parts +// array and per mesh object; handing out the held objects kept all of that alive with them +// (roadster_v10: 2.7 GiB after its shot, against 77 MiB of vertex arrays). A view's derived data +// goes when the command drops the view. The oldest compiles go once the arrays pass `limitBytes`. + +import { type Asset, type CompiledAsset, type CompileOptions, compile } from '../core/index.ts'; + +/** Vertex-array bytes held before the oldest compile is dropped (roadster_v10's six: 77 MiB). */ +export const SHARED_COMPILE_BYTES = 1024 ** 3; + +interface Held { + readonly asset: Asset; + readonly key: string; + readonly bytes: number; +} + +interface Shared { + readonly limit: number; + readonly byAsset: Map>; + /** Insertion order, oldest first. */ + readonly held: Held[]; + bytes: number; +} + +let shared: Shared | undefined; + +/** Bytes of every typed array under `v` (meshes, morphs, textures); a shared buffer counts twice. */ +function bytesOf(v: unknown, depth = 0): number { + if (ArrayBuffer.isView(v)) return v.byteLength; + if (depth > 6 || v === null || typeof v !== 'object') return 0; + let n = 0; + for (const x of Object.values(v)) n += bytesOf(x, depth + 1); + return n; +} + +/** What tells two compiles of one asset apart. */ +export const compileKey = (asset: Asset, o: CompileOptions): string => + JSON.stringify([o.lod ?? 0, o.seed ?? asset.seed ?? 1, o.variant ?? null]); + +/** + * Turn sharing on for this process; returns `stop`, which turns it off and drops every compile + * held. Only for a process whose recipe files, neural outputs and caches do not change meanwhile. + */ +export function shareCompiles(limitBytes: number = SHARED_COMPILE_BYTES): () => void { + const mine: Shared = { limit: limitBytes, byAsset: new Map(), held: [], bytes: 0 }; + shared = mine; + return () => { + if (shared === mine) shared = undefined; + }; +} + +/** How many compiles are held and their mesh bytes (0, 0 when sharing is off). */ +export const sharedCompiles = (): { readonly count: number; readonly bytes: number } => ({ + count: shared?.held.length ?? 0, + bytes: shared?.bytes ?? 0, +}); + +function drop(s: Shared): void { + const old = s.held.shift(); + if (!old) return; + s.bytes -= old.bytes; + const of = s.byAsset.get(old.asset); + of?.delete(old.key); + if (of?.size === 0) s.byAsset.delete(old.asset); +} + +/** New asset, parts and mesh objects over the same typed arrays: nothing memoised per object leaks. */ +const viewOf = (a: CompiledAsset): CompiledAsset => ({ + ...a, + parts: a.parts.map((p) => (p.mesh ? { ...p, mesh: { ...p.mesh } } : { ...p })), +}); + +/** + * `compile(asset, options)`; while sharing is on, a view of the compile this process already made + * of the same thing. + */ +export function compileOnce(asset: Asset, options: CompileOptions = {}): CompiledAsset { + const s = shared; + if (!s || options.tuned !== undefined) return compile(asset, options); + const key = compileKey(asset, options); + const hit = s.byAsset.get(asset)?.get(key); + if (hit) return viewOf(hit); + const made = compile(asset, options); + const bytes = bytesOf(made.parts); + // The newest compile is always kept, even alone over the limit: it is about to be used again. + while (s.held.length > 0 && s.bytes + bytes > s.limit) drop(s); + const of = s.byAsset.get(asset) ?? new Map(); + of.set(key, made); + s.byAsset.set(asset, of); + s.held.push({ asset, key, bytes }); + s.bytes += bytes; + return viewOf(made); +} diff --git a/src/cli/critique-cmd.ts b/src/cli/critique-cmd.ts index c6f3ebee..a1bfc75c 100644 --- a/src/cli/critique-cmd.ts +++ b/src/cli/critique-cmd.ts @@ -7,7 +7,7 @@ import { mkdir } from 'node:fs/promises'; import { dirname, join, relative, resolve } from 'node:path'; -import { type Asset, type CompiledAsset, compile, encodePng, type Lod } from '../core/index.ts'; +import { type Asset, type CompiledAsset, encodePng, type Lod } from '../core/index.ts'; import { DEFAULT_SHEET, type PaintedSheet, paintTrimSheet } from '../material/index.ts'; import { ART_RENDER, @@ -48,6 +48,7 @@ import { renderBeauty, } from '../render/index.ts'; import type { Args } from './args.ts'; +import { compileOnce } from './compile-cache.ts'; import type { RenderResult } from './render-cmds.ts'; export const sha256 = (b: Uint8Array): string => @@ -96,8 +97,8 @@ export function critiqueImages( asset: Asset, o: BeautyOptions, ): { images: Map; lods: CompiledAsset[]; metrics: ArtMetrics } { - const lods = ([0, 1, 2] as Lod[]).map((lod) => compile(asset, { lod })); - const lod0 = lods[0] ?? compile(asset); + const lods = ([0, 1, 2] as Lod[]).map((lod) => compileOnce(asset, { lod })); + const lod0 = lods[0] ?? compileOnce(asset); const lod2 = lods[2] ?? lod0; const images = new Map(); const hero = renderBeauty(lod0, o); @@ -218,7 +219,7 @@ export async function critiqueOne( const dir = join(args.out, asset.name, 'critique'); await mkdir(dir, { recursive: true }); const { images, lods, metrics } = critiqueImages(asset, { ...args.render, sheet }); - const lod0 = lods[0] ?? compile(asset); + const lod0 = lods[0] ?? compileOnce(asset); // The bound hero is the committed one: path traced at the shots setting (render/path/). images.set('hero.png', await pathHeroThreaded(lod0, { ...args.render, ...HERO_PATH, sheet })); const renders = Object.fromEntries([...images].map(([n, b]) => [n, sha256(b)])); diff --git a/src/cli/gaps-cmd.ts b/src/cli/gaps-cmd.ts index 17244ca0..529f39e1 100644 --- a/src/cli/gaps-cmd.ts +++ b/src/cli/gaps-cmd.ts @@ -6,7 +6,7 @@ import { mkdir } from 'node:fs/promises'; import { join } from 'node:path'; -import { type Asset, compile } from '../core/index.ts'; +import type { Asset } from '../core/index.ts'; import { describeGap, gapFinding, @@ -16,6 +16,7 @@ import { worstPerPair, } from '../qa/index.ts'; import type { Args } from './args.ts'; +import { compileOnce } from './compile-cache.ts'; import type { RenderResult } from './render-cmds.ts'; const ZOOMS = 6; @@ -23,7 +24,7 @@ const ZOOMS = 6; export async function gapsOne(asset: Asset, args: Pick): Promise { const dir = join(args.out, asset.name, 'gaps'); await mkdir(dir, { recursive: true }); - const scan = scanGaps(compile(asset, { lod: 0 }), { includeAllowed: true, overlayAll: true }); + const scan = scanGaps(compileOnce(asset, { lod: 0 }), { includeAllowed: true, overlayAll: true }); const files = [join(dir, 'overlay.png')]; await Bun.write(files[0] ?? '', gapOverlay(scan, 500)); const open = worstPerPair(scan.gaps.filter((g) => !g.allowed)); diff --git a/src/cli/icon-cmd.ts b/src/cli/icon-cmd.ts index 3e5e5914..86be06d0 100644 --- a/src/cli/icon-cmd.ts +++ b/src/cli/icon-cmd.ts @@ -5,10 +5,11 @@ import { mkdir } from 'node:fs/promises'; import { join } from 'node:path'; -import { type Asset, compile, type ErrorCode, type Lod } from '../core/index.ts'; +import type { Asset, ErrorCode, Lod } from '../core/index.ts'; import type { PaintedSheet } from '../material/index.ts'; import { icon, portrait, type ReviewWarning, rtsView, teamCoverage } from '../render/index.ts'; import type { Args } from './args.ts'; +import { compileOnce } from './compile-cache.ts'; import { type RenderResult, variantOf } from './render-cmds.ts'; /** Team colour must cover at least this share of the 64-px RTS silhouette. */ @@ -40,7 +41,7 @@ export async function iconOne( await mkdir(dir, { recursive: true }); const variant = variantOf(asset, args); const lod: Lod = 0; - const lod0 = compile(asset, { lod, ...(variant ? { variant } : {}) }); + const lod0 = compileOnce(asset, { lod, ...(variant ? { variant } : {}) }); const png = (n: string): string => join(dir, variant ? `${n}_${variant}.png` : `${n}.png`); const focus = focusOf(asset); const files: [string, Uint8Array][] = [ @@ -53,8 +54,8 @@ export async function iconOne( const pair = teamPair(asset); if (pair) { const c = teamCoverage( - compile(asset, { lod: 2, variant: pair[0] }), - compile(asset, { lod: 2, variant: pair[1] }), + compileOnce(asset, { lod: 2, variant: pair[0] }), + compileOnce(asset, { lod: 2, variant: pair[1] }), { sheet }, ); if (c.share < MIN_TEAM_SHARE) diff --git a/src/cli/neural-store.test.ts b/src/cli/neural-store.test.ts new file mode 100644 index 00000000..0065e84a --- /dev/null +++ b/src/cli/neural-store.test.ts @@ -0,0 +1,60 @@ +// provideRecipeNeural's miss text depends on where the recipe sits: a checkout says "run +// `t3d neural run` and commit", an installed package (which ships no recipes/*/neural/) says where +// the model outputs are. Both the notice and the `--require-neural` error carry it. + +import { afterAll, describe, expect, it } from 'bun:test'; +import { mkdir, rm } from 'node:fs/promises'; +import { join } from 'node:path'; +import { isTs23dError } from '../core/index.ts'; +import { provideRecipeNeural } from './neural-store.ts'; + +const dir = join(import.meta.dir, '../../tmp/cli-neural-store-test'); +afterAll(() => rm(dir, { recursive: true, force: true })); + +const REQ = { seed: 1, prompt: 'lichen', width: 4, height: 4 } as const; +const installed = join(dir, 'app/node_modules/@developerz.ai/ts23d/recipes/bear/bear.ts'); +const checkout = join(dir, 'repo/recipes/bear/bear.ts'); + +const thrown = (run: () => unknown): unknown => { + try { + return run(); + } catch (e) { + return e; + } +}; + +describe('provideRecipeNeural: the fix of a miss', () => { + it('installed package without neural/: the notice says the package ships no model outputs', async () => { + const cache = await provideRecipeNeural(installed, ['bear'], false); + const r = cache.get('image', REQ); + const fix = r.fallback ? r.notice.fix : ''; + expect(fix).toContain('npm package does not ship model outputs'); + expect(fix).toContain('https://github.com/developerz-ai/ts23d'); + expect(fix).toContain('copy the recipe folder'); + expect(fix).toContain('t3d neural run'); + expect(fix).not.toContain('then commit'); + }); + + it('installed package, --require-neural: the error carries the same fix', async () => { + const cache = await provideRecipeNeural(installed, ['bear'], true); + const e = thrown(() => cache.get('image', REQ)); + expect(isTs23dError(e) && e.code).toBe('T3D_NEURAL_MISSING'); + expect(isTs23dError(e) && e.fix).toContain('npm package does not ship model outputs'); + }); + + it('a checkout keeps the run-and-commit fix', async () => { + const r = (await provideRecipeNeural(checkout, ['bear'], false)).get('image', REQ); + expect(r.fallback && r.notice.fix).toContain('t3d neural run recipes/bear/bear.ts'); + expect(r.fallback && r.notice.fix).toContain('then commit recipes/bear/neural/'); + }); + + it('a package folder that does hold neural/ (a linked checkout) keeps the run-and-commit fix', async () => { + const linked = join(dir, 'linked/node_modules/@developerz.ai/ts23d/recipes/bear'); + await mkdir(join(linked, 'neural'), { recursive: true }); + const r = (await provideRecipeNeural(join(linked, 'bear.ts'), ['bear'], false)).get( + 'image', + REQ, + ); + expect(r.fallback && r.notice.fix).toContain('then commit'); + }); +}); diff --git a/src/cli/neural-store.ts b/src/cli/neural-store.ts index 30fd0fbe..25206354 100644 --- a/src/cli/neural-store.ts +++ b/src/cli/neural-store.ts @@ -5,7 +5,7 @@ // Local outputs (`local-.*`, kept off git) are read only by `t3d neural run`, which sends them // to the next request; a build never reads them, so it is the same build on every machine. -import { mkdir, readdir, readFile, writeFile } from 'node:fs/promises'; +import { mkdir, readdir, readFile, stat, writeFile } from 'node:fs/promises'; import { basename, dirname, join, resolve } from 'node:path'; import { entryOf, @@ -79,6 +79,18 @@ async function ignoreLocal(dir: string): Promise { ); } +const INSTALLED = '/node_modules/@developerz.ai/ts23d/'; + +/** + * The fix of a miss for a recipe inside an installed package with no `neural/` folder: the npm + * package ships recipes without their model outputs, so "run it, then commit" is the wrong advice. + */ +async function installedFix(recipe: string, dir: string): Promise { + if (!dir.replaceAll('\\', '/').includes(INSTALLED)) return undefined; + if (await stat(dir).catch(() => undefined)) return undefined; + return `the npm package does not ship model outputs (they stay in git): clone https://github.com/developerz-ai/ts23d and build recipes/${recipe}/${recipe}.ts there, or copy the recipe folder into your project and run \`t3d neural run /${recipe}.ts\` on a machine with the adapters its README names`; +} + /** Load the recipe's committed cache and provide it to compiles of `assets` (offline). */ export async function provideRecipeNeural( recipeFile: string, @@ -86,10 +98,12 @@ export async function provideRecipeNeural( requireNeural: boolean, ): Promise { const { recipe, dir } = neuralDirOf(recipeFile); + const missingFix = await installedFix(recipe, dir); const cache = neuralCache({ recipe, offline: true, requireNeural, + ...(missingFix ? { missingFix } : {}), entries: await loadNeuralEntries(dir, { local: false }), }); provideNeural(cache, assets); diff --git a/src/cli/render-cmds.ts b/src/cli/render-cmds.ts index 6e06a81e..5ce9d4e0 100644 --- a/src/cli/render-cmds.ts +++ b/src/cli/render-cmds.ts @@ -6,7 +6,7 @@ import { mkdir } from 'node:fs/promises'; import { dirname, join, relative, resolve } from 'node:path'; -import { type Asset, compile, Ts23dError } from '../core/index.ts'; +import { type Asset, Ts23dError } from '../core/index.ts'; import type { PaintedSheet } from '../material/index.ts'; import { poseFrames } from '../qa/index.ts'; import { @@ -18,6 +18,7 @@ import { reviewPack, } from '../render/index.ts'; import type { Args } from './args.ts'; +import { compileOnce } from './compile-cache.ts'; import { grantedCpu } from './queue/index.ts'; export interface RenderResult { @@ -51,7 +52,7 @@ export async function heroOne( await mkdir(dir, { recursive: true }); const variant = variantOf(asset, args); const path = join(dir, variant ? `hero_${variant}.png` : 'hero.png'); - const lod0 = compile(asset, { lod: 0, ...(variant ? { variant } : {}) }); + const lod0 = compileOnce(asset, { lod: 0, ...(variant ? { variant } : {}) }); // Under the queue the tracer uses the CPU tokens it was granted (same bytes at any count). const threads = grantedCpu(); await Bun.write( @@ -73,7 +74,7 @@ export async function blueprintOne(asset: Asset, args: Args): Promise 0) out.set('poses.png', poseStrip(lod1, frames)); diff --git a/src/core/neural-ctx.ts b/src/core/neural-ctx.ts index 41e5b21c..bcb6fd3e 100644 --- a/src/core/neural-ctx.ts +++ b/src/core/neural-ctx.ts @@ -28,6 +28,12 @@ export interface NeuralCtxAsk { * the hit's outputs are `absent`: for raw meshes too big to commit that only feed a next request. */ readonly keep?: 'local'; + /** + * The asset has no CPU path worth building (a model-first body): in a build, a miss of this ask + * stops with T3D_NEURAL_MISSING, not `null`. `t3d neural run` still lists it as a request to + * fill, and a compile with no cache provided at all still gets `null`. + */ + readonly required?: boolean; } /** Where cached outputs come from (/neural's recipe cache implements it). */ diff --git a/src/kinds/model-chain.test.ts b/src/kinds/model-chain.test.ts index b0519951..ac8e8aed 100644 --- a/src/kinds/model-chain.test.ts +++ b/src/kinds/model-chain.test.ts @@ -50,7 +50,11 @@ function fake(outputs: Partial>): { return { source, seen }; } -function run(name: string, source: NeuralSource): ImportedModel | null | undefined { +function run( + name: string, + source: NeuralSource, + required = false, +): ImportedModel | null | undefined { let got: ChainedModel | null | undefined; const asset = defineAsset({ name, @@ -65,6 +69,7 @@ function run(name: string, source: NeuralSource): ImportedModel | null | undefin models: { mesh: 'pixal3d' }, license: 'MIT', author: 'fixture', + ...(required ? { required } : {}), }); }, }); @@ -97,6 +102,19 @@ describe('modelChain', () => { expect(noBlend.seen.map((s) => s.task)).toEqual(['image', 'mesh', 'blender']); }); + it('required: every ask says so (a build cache then stops at the first miss); default: none', () => { + const plain = fake({ image, mesh: absentMesh, blender }); + run('chain_optional', plain.source); + expect(plain.seen.map((s) => s.ask?.required)).toEqual([undefined, undefined, undefined]); + const f = fake({ image, mesh: absentMesh, blender }); + run('chain_required', f.source, true); + expect(f.seen.map((s) => [s.task, s.ask?.required])).toEqual([ + ['image', true], + ['mesh', true], + ['blender', true], + ]); + }); + it('returns null when the Blender output itself is absent (bytes are needed to import)', () => { const f = fake({ image, diff --git a/src/kinds/model-chain.ts b/src/kinds/model-chain.ts index 6709e55e..6b0938a8 100644 --- a/src/kinds/model-chain.ts +++ b/src/kinds/model-chain.ts @@ -12,7 +12,9 @@ // are on this machine, so `t3d neural run` can send them (one round per ask: five with a rigger). The fit is // recomputed from the committed proxy at every build, by the same code that produced the // program. A plain `steps` array skips the proxy: one pass, registered in TS on the output (for -// rigged models, `clips: 'model'`). Any miss returns null: the CPU creature. +// rigged models, `clips: 'model'`). Any miss returns null: the CPU creature. With `required`, a +// miss in a build is T3D_NEURAL_MISSING (the cache throws it): for a recipe whose CPU creature is +// not an asset it would ship. import { type Bone, @@ -89,6 +91,12 @@ export interface ModelChainOptions { /** Seed of the Blender request. Default: the surface request's. */ readonly seed?: number; readonly import?: ImportModelOptions; + /** + * The model is the asset: `t3d build` with a missing output stops with one T3D_NEURAL_MISSING + * instead of building the code creature. Default off (a miss returns null). A compile with no + * cache provided (a test, the gallery's CPU tier) still returns null. + */ + readonly required?: boolean; } /** The output's content hash: the cache's, else of the bytes in hand; null when neither. */ @@ -116,7 +124,10 @@ export interface ChainedModel { /** The imported Blender-steps model, or null on any cache miss (build the code creature). */ export function modelChain(ctx: BuildContext, o: ModelChainOptions): ChainedModel | null { - const pin = (model: string | undefined) => (model === undefined ? {} : { model }); + const pin = (model: string | undefined) => ({ + ...(model === undefined ? {} : { model }), + ...(o.required ? { required: true } : {}), + }); const png = ctx.neural('image', o.concept, pin(o.models?.image))?.outputs[0]; const image = hashOf(png); if (!image) return null; @@ -182,7 +193,7 @@ export function modelChain(ctx: BuildContext, o: ModelChainOptions): ChainedMode const rigged = ctx.neural( 'rig', { seed, mesh: { sha256: sha, mime: 'model/gltf-binary' } }, - { model: o.rig, inputs: new Map([[sha, proxy]]) }, + { ...pin(o.rig), inputs: new Map([[sha, proxy]]) }, )?.outputs[0]; if (!present(rigged)) return null; joints = importModel(source(rigged.bytes, 'rig'), { place: 'origin' }).armature?.bones; diff --git a/src/kinds/model-joints.test.ts b/src/kinds/model-joints.test.ts index aa81ec46..f08471a9 100644 --- a/src/kinds/model-joints.test.ts +++ b/src/kinds/model-joints.test.ts @@ -36,6 +36,54 @@ describe('rigTopology', () => { expect(t.limbs.hind_l[0]?.head).toEqual([-0.28, 0.08, 0.5]); }); + // D293: UniRig may end a leg in a millimetre-long stub bone past the toe. + /** The fixture's hind legs (its last eight bones), each with one more bone `long` metres long. */ + const withHindEnds = (long: number): Bone[] => { + const rig = riggerBones(); + for (const leaf of [rig[rig.length - 5], rig[rig.length - 1]]) + if (leaf) + rig.push({ + name: `bone_${rig.length}`, + head: leaf.tail, + tail: [leaf.tail[0] + long, leaf.tail[1], leaf.tail[2]], + parent: leaf.name, + }); + return rig; + }; + + it('drops a 6 mm terminal stub from a five-bone hind leg: hip → toe are the four real bones', () => { + const t = rigTopology(withHindEnds(0.006)); + const real = rigTopology(riggerBones()); + expect(t.limbs.hind_l.map((b) => b.name)).toEqual(real.limbs.hind_l.map((b) => b.name)); + expect(t.limbs.hind_r.map((b) => b.name)).toEqual(real.limbs.hind_r.map((b) => b.name)); + expect(t.limbs.hind_l[0]?.head).toEqual([-0.28, 0.08, 0.5]); + // The stub is still part of the leg: it is not taken for the head or a tail. + expect(t.head.name).toBe(real.head.name); + expect(t.tail).toBeUndefined(); + const mapped = mapJoints(plan, withHindEnds(0.006), () => true); + expect(mapped.find((b) => b.name === 'hind_l_0')?.head).toEqual([-0.28, 0.08, 0.5]); + expect(mapped.find((b) => b.name === 'hind_r_3')?.tail).toEqual([-0.2, -0.08, 0.01]); + }); + + it('keeps a four-bone leg whole even when its last bone is 6 mm: there is no bone to spare', () => { + const rig = riggerBones(); + const leaf = rig[rig.length - 1]; + if (leaf) rig[rig.length - 1] = { ...leaf, tail: [leaf.head[0] + 0.006, -0.08, leaf.head[2]] }; + const t = rigTopology(rig); + expect(t.limbs.hind_r.map((b) => b.name)).toEqual(rig.slice(-4).map((b) => b.name)); + expect(t.limbs.hind_r[0]?.head).toEqual([-0.28, -0.08, 0.5]); + }); + + it('keeps a short real toe: 0.139 × the mean of the other bones, the shortest in the committed rigs', () => { + // The fixture's four leg bones average 0.133 m: 0.0185 m is the bear's toe-to-leg ratio. + const rig = withHindEnds(0.0185); + const t = rigTopology(rig); + // rig ends: … hind_l ×4, hind_r ×4, left toe, right toe. The last four of each five stay. + const names = (at: number[]): string[] => at.map((k) => rig[rig.length + k]?.name ?? ''); + expect(t.limbs.hind_l.map((b) => b.name)).toEqual(names([-9, -8, -7, -2])); + expect(t.limbs.hind_r.map((b) => b.name)).toEqual(names([-5, -4, -3, -1])); + }); + it('finds the trunk, the head and the four limbs by shape, not by name', () => { const t = rigTopology(riggerBones()); expect(t.trunk.map((b) => b.name)).toEqual([ diff --git a/src/kinds/model-joints.ts b/src/kinds/model-joints.ts index 84838287..62de64b9 100644 --- a/src/kinds/model-joints.ts +++ b/src/kinds/model-joints.ts @@ -3,8 +3,8 @@ // the asset frame): // limbs — the four leaf chains of four or more bones that end on the ground; their last four // bones are hip / shoulder → toe, a fifth above them is the scapula (UniRig gives a -// wolf's fore legs five). The pair hanging from further along the trunk is the fore -// pair, +y is left +// wolf's fore legs five). A terminal stub is dropped first (STUB_SHARE). The pair +// hanging from further along the trunk is the fore pair, +y is left // head — the front-most leaf that is not a limb, or its parent when that leaf is one of a // cluster of leaves (ears and jaw hanging from the skull bone) // trunk — the path from the root to the head bone; the plan's spine and neck joints are laid on @@ -34,6 +34,21 @@ export const RIG_CHAINS: readonly RigChains[] = ['trunk', 'fore', 'hind']; /** Share of the rig's height a limb's last bone must end within, from the lowest point. */ const GROUND_SHARE = 0.1; +/** A limb's last bone shorter than this share of the mean length of the chain's other bones is a + * rigger's stub, not a toe (UniRig: a 6 mm fifth bone on a hind leg, which put the hip on the + * stifle; 6 mm is 4–9 % of that mean on the committed rigs' legs). Their shortest real toe is + * 14 % (bear_hybrid), and 26 % among chains with a bone to spare (dragon_hybrid). */ +const STUB_SHARE = 0.1; + +const long = (b: Bone): number => v3.len(v3.sub(b.tail, b.head)); + +/** A limb chain without terminal stubs; a chain of four is kept whole (no bone to spare). */ +const unstubbed = (c: Bone[]): Bone[] => { + const [rest, last] = [c.slice(0, -1), c[c.length - 1]]; + const mean = rest.reduce((t, b) => t + long(b), 0) / rest.length; + return c.length > 4 && last && long(last) < mean * STUB_SHARE ? unstubbed(rest) : c; +}; + export interface RigTopology { /** Root → head bone's parent. */ readonly trunk: readonly Bone[]; @@ -81,7 +96,9 @@ export function rigTopology(source: readonly Bone[]): RigTopology { fail( `${full.length} four-bone limb chain(s) reaching the ground found, a quadruped has 4 (leaf chains: ${chains.map((c) => c.length).join(', ')})`, ); + // The stubs stay in `inLeg`: a dropped one is still no head and no tail. const inLeg = new Set(full.flat().map((b) => b.name)); + const legs = full.map(unstubbed); const front = leaves.filter((b) => !inLeg.has(b.name)).sort((p, q) => q.tail[0] - p.tail[0])[0]; if (!front) return fail('no head bone: every leaf belongs to a limb'); // A leaf among leaves (ears, jaw) hangs from the skull bone: that one is the head. @@ -92,8 +109,8 @@ export function rigTopology(source: readonly Bone[]): RigTopology { const trunk: Bone[] = []; for (let p = byName.get(head.parent ?? ''); p; p = byName.get(p.parent ?? '')) trunk.unshift(p); const at = (c: readonly Bone[]): number => trunk.findIndex((b) => b.name === c[0]?.parent); - if (full.some((c) => at(c) < 0)) fail('a limb does not hang from the trunk (root → head path)'); - const ordered = [...full].sort( + if (legs.some((c) => at(c) < 0)) fail('a limb does not hang from the trunk (root → head path)'); + const ordered = [...legs].sort( (p, q) => at(p) - at(q) || (q[0]?.head[0] ?? 0) - (p[0]?.head[0] ?? 0), ); const [h1, h2, f1, f2] = ordered; diff --git a/src/neural/recipe-cache.test.ts b/src/neural/recipe-cache.test.ts index 70e063e9..b8d3f72d 100644 --- a/src/neural/recipe-cache.test.ts +++ b/src/neural/recipe-cache.test.ts @@ -76,6 +76,35 @@ describe('recipe cache: misses (failure first)', () => { const e = await c.fill(mockAdapter(), job('mesh')).catch((x: unknown) => x); expect(e instanceof Ts23dError && e.code).toBe('T3D_NEURAL_MISSING'); }); + test('missingFix replaces the checkout fix in the notice and in the error', () => { + const fix = 'the npm package does not ship model outputs'; + const c = neuralCache({ recipe: 'oak', offline: true, missingFix: fix }); + const r = c.get('image', REQ.image); + expect(r.fallback && r.notice.fix).toBe(fix); + const strict = neuralCache({ + recipe: 'oak', + offline: true, + requireNeural: true, + missingFix: fix, + }); + const e = (() => { + try { + return strict.get('image', REQ.image); + } catch (x) { + return x; + } + })(); + expect(e instanceof Ts23dError && [e.code, e.fix]).toEqual(['T3D_NEURAL_MISSING', fix]); + }); + test('a required ask: a build (offline) throws on its miss, a filling cache lists it', () => { + const build = neuralCache({ recipe: 'bear', offline: true }); + expect(() => build.get('image', REQ.image, { required: true })).toThrow('T3D_NEURAL_MISSING'); + expect(() => build.get('image', REQ.image, { required: true })).not.toThrow('CPU path'); + expect(build.notices()).toEqual([]); + const fill = neuralCache({ recipe: 'bear' }); + expect(fill.get('image', REQ.image, { required: true }).fallback).toBe(true); + expect(fill.misses()).toEqual([{ task: 'image', request: REQ.image }]); + }); test('over the cap (12 MB default) → T3D_NEURAL_FAILED, nothing stored', async () => { const c = neuralCache({ recipe: 'big', capBytes: 64 }); const e = await c.fill(mockAdapter(), job('image')).catch((x: unknown) => x); diff --git a/src/neural/recipe-cache.ts b/src/neural/recipe-cache.ts index a711551b..305995ad 100644 --- a/src/neural/recipe-cache.ts +++ b/src/neural/recipe-cache.ts @@ -3,8 +3,8 @@ // synchronous on the read side, like the import cache: the CLI loads the directory (src/cli/ // neural-store.ts) and hands it over with `provideNeural`; a compile asks `providedNeural(recipe) // .get(task, request)`. A miss returns `{ fallback: true }` plus a T3D_NEURAL_MISSING notice so the -// recipe takes its CPU path — an error only with `requireNeural`. `fill` runs an adapter (never in -// a build) and validates before anything is stored. `provideNeural` also serves `ctx.neural`. +// recipe takes its CPU path — an error with `requireNeural`, or for a `required` ask in a build +// (offline). `fill` runs an adapter (never in a build) and validates before anything is stored. `provideNeural` also serves `ctx.neural`. // A `keep: 'local'` ask stores **local** outputs (`local-.*`, git-ignored, outside the cap): a // machine without the file still gets a hit, its outputs `absent` with only their sha256, so a // chain (image → raw mesh → blender) reaches the committed downstream output. @@ -55,6 +55,8 @@ export interface NeuralAsk { readonly inputs?: ReadonlyMap; /** Storage policy of a fill, never part of the key: outputs stay on this machine. */ readonly keep?: 'local'; + /** No CPU path for this ask: its miss is an error in an offline (build) cache. */ + readonly required?: boolean; } /** A job plus where its outputs are kept. */ @@ -131,6 +133,8 @@ export function neuralCache(o: { readonly offline?: boolean; /** Bytes allowed across all committed outputs; default NEURAL_CAP_BYTES (ADR 014). */ readonly capBytes?: number; + /** The `fix:` of a miss where "run it, then commit" does not apply (an installed package). */ + readonly missingFix?: string; }): RecipeNeuralCache { const cap = o.capBytes ?? NEURAL_CAP_BYTES; const byKey = new Map(); @@ -139,11 +143,20 @@ export function neuralCache(o: { const run = `t3d neural run recipes/${o.recipe}/${o.recipe}.ts`; const total = (): number => [...byKey.values()].reduce((s, e) => s + size(e), 0); - const miss = (job: NeuralCacheJob, rk: string): NeuralNotice => { + const miss = (job: NeuralCacheJob, rk: string, required: boolean): NeuralNotice => { const { task, model } = job; - const message = `${o.recipe}: no cached ${task} output for request ${rk.slice(0, 12)}…${model ? ` (model ${model})` : ''}; using the CPU path`; - const fix = `run \`${run}\` on a machine with a ${task} adapter (t3d.config.ts), then commit recipes/${o.recipe}/neural/`; - if (o.requireNeural) throw new Ts23dError('T3D_NEURAL_MISSING', { cause: message, fix }); + const what = `${o.recipe}: no cached ${task} output for request ${rk.slice(0, 12)}…${model ? ` (model ${model})` : ''}`; + const fix = + o.missingFix ?? + `run \`${run}\` on a machine with a ${task} adapter (t3d.config.ts), then commit recipes/${o.recipe}/neural/`; + if (o.requireNeural) throw new Ts23dError('T3D_NEURAL_MISSING', { cause: what, fix }); + // A model-first asset in a build: one error, not its CPU stand-in's QA findings. + if (required && o.offline) + throw new Ts23dError('T3D_NEURAL_MISSING', { + cause: `${what}, and the asset is model-first: it has no CPU build`, + fix, + }); + const message = `${what}; using the CPU path`; const notice: NeuralNotice = { code: 'T3D_NEURAL_MISSING', message, fix }; notices.set(`${rk}:${model ?? ''}`, { notice, job }); return notice; @@ -168,7 +181,7 @@ export function neuralCache(o: { ...(ask.inputs ? { inputs: ask.inputs } : {}), ...(ask.keep ? { keep: ask.keep } : {}), }; - return { fallback: true, notice: miss(job, rk) }; + return { fallback: true, notice: miss(job, rk, ask.required === true) }; }; return {