docs(plan): the website program is complete in-repo; two checks wait on a tag

Six of seven "done when" criteria verified today. The seventh — /reference
showing the metamodel — cannot be checked until a tag carries site-reference/,
which v0.24.5 predates.

Records what the program added that the plan did not ask for, since all three
came from one constraint the plan underestimated: the deploy pins to a release
tag and installs nothing. That is why the reference renders here rather than at
deploy, why the llms mirrors are gated on the tag, and why the tag itself had to
stop being cut halfway through a coordinated release.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NynrRND6ZUwGvq3ZUTCfxG

Author: dmealing

docs/superpowers/plans/2026-08-29-website-self-updating-codegen.md

@@ -2401,10 +2401,35 @@ Any nav link labelled "Reference" now points at the metamodel. Add a link to the
 
 ## Done when
 
-- Every metadata and generated-code block on `metaobjects.dev` comes from `site-payload.json`.
-- `bash scripts/ci-local.sh --only gates` is green, including `gate_site_payload` — and that lane still needs only `bun`, still runs offline, and still does not write to the working tree.
-- The release preflight passes, verified via `--preflight-only` (**never** via `release.mjs <v> --dry-run`, which bumps 14 package.json files and regenerates the lockfile before it exits).
-- Each of the eight gates has been seen to FAIL when its condition is broken, and then to pass again.
-- The live site renders every block, `<details>` expands, and `view-source` shows no `<script>`.
-- `/reference` shows the metamodel — all 69 `type.subType` pairs — and `/reference/example` shows the adopter model. No nav link 404s.
-- `meta docs --metamodel --site` refuses with a message instead of silently dropping `--site`.
+- [x] Every metadata and generated-code block on `metaobjects.dev` comes from `site-payload.json`.
+- [x] `bash scripts/ci-local.sh --only gates` is green, including `gate_site_payload` — and that lane still needs only `bun`, still runs offline, and still does not write to the working tree.
+- [x] The release preflight passes, verified via `--preflight-only` (**never** via `release.mjs <v> --dry-run`, which bumps 14 package.json files and regenerates the lockfile before it exits).
+- [x] Each of the eight gates has been seen to FAIL when its condition is broken, and then to pass again.
+- [x] The live site renders every block, `<details>` expands, and `view-source` shows no `<script>`.
+- [~] `/reference` shows the metamodel — all 69 `type.subType` pairs — and `/reference/example` shows the adopter model. No nav link 404s.
+- [x] `meta docs --metamodel --site` refuses with a message instead of silently dropping `--site`.
+
+> **STATUS 2026-08-30 — the program is complete in this repo; two live verifications wait
+> on the next release.**
+>
+> Verified today: 15 placeholders live on metaobjects.dev, all filled, zero `<script>`, 8
+> `<details>` expanders; the `gates` lane green at 20 checks (now including `site reference
+> is fresh`), offline, bun-only, leaving the tree clean; the release preflight green via
+> `--preflight-only`; `site-reference/index.html` carries exactly **69** `type.subType`
+> pairs.
+>
+> **The one criterion not yet met live is `/reference`**, and only because `v0.24.5`
+> predates `site-reference/`. Both held site commits (`ef743d7` llms mirrors, `55f9aae`
+> reference + example) go out immediately after the next release, which `finish-release.mjs`
+> now guarantees will carry both the renderer output and fresh llms mirrors.
+>
+> **What this program added that the plan did not ask for, all forced by the same
+> constraint** — the deploy pins to a release tag and installs nothing:
+> - `scripts/finish-release.mjs`, because the tag was cut in the MIDDLE of a coordinated
+>   release and therefore named a tree stating three registry versions that were never
+>   published. The website reads that tree. Fixed as ORDER, gated six ways, proven by
+>   breaking each gate.
+> - `site-reference/` rendered here rather than at deploy, and gated for freshness on both
+>   difference and orphan.
+> - The llms freshness gate on the tag, for the same reason: the site copies from the tag,
+>   and the docs refresh had always been a post-tag commit.