Repository navigation
Land profile calibration and selection outcomes - #49
Conversation
An autonomous writer needs somewhere to get its facts and somewhere its output can be checked without a human reading every letter. This adds both, and keeps them apart: the fact ledger is objective and gated on human verification, the style card is subjective and learned from corrections. A style mistake is recoverable; an unsupported claim sent under the applicant's name is not. Facts are proposed by local resume extraction and are unusable until verified, with an optional expiry for anything that goes stale. Drafts are logged through a new WebMCP tool at any volume, but every sentence asserting something checkable must cite a verified fact, and every number must appear in a cited fact. A draft failing either rule is refused at the API and never written, so an invented achievement cannot reach a review queue at all. Review is batched and ordered by risk rather than convenience: an unseen role cluster reviews at the first draft, then style drift, then low agent confidence, then a full batch. Sessions group repeated openings so one decision covers several drafts and turn corrections into style rules, since a session that changes no rule leaves the next batch identical. Autonomy is then earned per cluster. Enough near-unchanged reviews graduate a cluster into staging its own drafts; a correction or expired fact demotes it. Staging is not acceptance — status and accepted_draft are untouched, so accepting an exact draft stays a human action, and agents still cannot submit anything. Every draft records the profile version that produced it, without which a bad draft cannot be attributed to a bad agent rather than a stale profile. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The review screen is built to resist becoming a batched approval queue. Reading twelve letters in sequence is the same work as before, so it leads with repeated openings grouped into one decision, flags near-identical drafts, and rolls up the facts leaned on most. Corrections are generalised into proposed rules and confirmed as "always, or just here?", because most people cannot describe their own voice but recognise a violation instantly. Closing with an empty rule basket says plainly that the next batch will repeat the same habits. The profile screen keeps extraction and verification visibly separate: a pasted resume proposes candidates and writes nothing, and each line stays unusable until verified individually. Docs state what agents may and may not do, and that citation enforcement checks a claim against a verified fact rather than checking that the surrounding wording is true. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Drafting was never the largest lever. Which jobs you apply to is: reply odds vary by orders of magnitude with fit, while letter quality varies perhaps twofold. This adds the selection half. The objective is deliberately not the sum over applications. You accept one job, so the quantity that matters is the expected value of the best single offer — an order statistic, computed exactly rather than approximated by tier heuristics. That one change produces a spread of odds by itself: once a likely offer is held, further similar roles add almost nothing and a long shot starts winning the comparison. Selection is greedy on gain per minute under a stated attention budget, so no reach/match/floor ratio is hardcoded, and every row carries the reason it earned its place. Preferences supply the utility term and cannot be asked for directly. People cannot state trade-off weights in the abstract but choose between two concrete jobs instantly, so they are elicited by forced choice over real postings and fitted by pairwise logistic regression. Domain fit is computed from the person's own verified claims rather than learned as one weight per role family, because a dozen comparisons cannot determine a dozen weights. With no answers recorded the model reports no opinion instead of inventing one. Outcomes close the loop the state machine previously left open at Live loop. Terminal states are local records of what happened and are deliberately not valid import statuses, which leaves the existing import matrix untouched and means rediscovery can never reopen an ended application — asserted against the real upsert SQL as well as the TypeScript, since the two must agree. A submission requires a receipt. A false "submitted" is doubly corrupting: deduplication would stop you applying to a job you never applied to, and the response-rate posterior would count a send that never happened. Imported submissions are kept and shown but arrive unreceipted and are excluded from the rates. Rates are reported as intervals with their evidence count, and are descriptive only. The read plane pulls public Greenhouse and Lever boards onto the existing jobKey identity. It reads only, needs no credentials and no account, and every posting arrives Held. No write-plane adapters are included. Nothing here submits an application; recording a submission is the applicant reporting what they already did. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The preferences screen shows two real jobs side by side and asks which you would rather have, because that is a question people can answer accurately and instantly. It shows the fitted weights as shares of one whole so the implied trade-offs are visible and arguable, and says plainly that it has no opinion until told something. The plan screen presents the week rather than a queue: what to apply to, what it cost of the budget, and why each row earned its place. It reports what the same budget would buy under a plain value ranking, so the portfolio's advantage is shown rather than asserted, and it says when the value model is still too thin to lean on. The track screen carries both halves of what happens after sending. Reply rates are labelled descriptive and shown with their evidence count, live applications are flagged when they carry no receipt, and each one lists the exact claims it commits you to defending — which is the citation list, not a document anyone had to write. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Bring closed #26 and open #46 onto 9296d9d without dropping session expiry, dirty-editor navigation, null-URL packet identity, public-copy generation, or the frozen toolchain. Agent drafts stay gated on verified facts; selection maximises the best single offer; outcomes require a receipt and cannot be reopened by import. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
…eck. The new Chromium journey clicked Import research before the signed-in shell finished rendering, so the JSON field never appeared. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
…ities. Resolve #49 conflicts against #45 public-copy: preserve dual-builder credit and recruiter wording, then add fact-ledger, batch review, planning, board pulls, and receipted outcomes. Regenerate the managed README block. Open a job from the existing queue for the dirty-navigation Chromium check. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
sdcarlson
left a comment
There was a problem hiding this comment.
Request changes on exact3e081b4ffb24c207e78537dd5efe9f16491e8cee. This replacement retains all four reviewed PR46 blockers. Independent recheck reran the exact-source function reproductions; outcomes.ts, preferences route and outcomes route are unchanged from46:
- lib/outcomes.ts:76-83: Offer-to-Accepted still throws because Offer is treated as terminal.
- lib/domain.ts:148-150: Held-to-Accepted with empty draft/blocker still passes ordinary workspace validation, bypassing the outcome path.
- app/api/preferences/route.ts:76-80: delta:[] is accepted and stored; fitWeights([[]]) yields five NaN weights.
- app/api/outcomes/route.ts:83-129: unconditional outcome/event inserts commit even when the version-checked job update changes zero rows; success is returned. Gate all related writes on the successful version mutation and return409 on conflict.
Additional integration gate: scripts/ci/run-unit.mjs now excludes calibration.test.mjs and orchestration.test.mjs, but package scripts/workflow/run-integration do not execute these suites elsewhere. Run them against dedicated freshly migrated fixtures as part of required CI, rather than excluding the new route coverage entirely. Preserve existing checks and privacy/stale-write assertions.
Please fix these cases with regression evidence, reconcile current main, and run the complete required suite before requesting final review. The documented additive-migration scope does not itself validate endpoint behavior. PR46 should not be merged independently; its commits can be preserved until this corrected replacement is accepted.
CodeQL flagged chained unescaping and a script/style filter that missed closing tags like </script >. Strip tags first, then decode named and numeric entities in a single pass so double-encoded sequences stay text. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
CodeQL js/bad-tag-filter still treated </script\\s*> as incomplete because browsers also accept end tags like </script\\t\\n bar>. Drop script and style elements by scanning tag boundaries instead of filtering with a regexp. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
Offer remains terminal for import and editor status, but accepted, withdrawn, and declined outcomes can still be recorded from it. Held-to-Accepted is refused in ordinary workspace validation. Empty comparison vectors are rejected and cannot fit NaN weights. Outcome and event writes run only after a successful versioned job update and return 409 on conflict. Applicant-claim detection no longer treats every digit as a claim: employer-directed numbers such as "your 2024 post" do not require a citation. Calibration and orchestration now run in required CI on dedicated freshly migrated databases. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
|
Ready for re-review @sdcarlson |
sdcarlson
left a comment
There was a problem hiding this comment.
Request changes on exact 8466042. The new CI matrix now executes calibration and orchestration on dedicated fixtures, and the empty-vector/transition cases have substantive fixes. Two exact-source reproductions still block approval:
-
lib/utility.ts isChoiceDelta validates length and finiteness but does not bound components. isChoiceDelta([1e308,1e308,1e308,1e308,1e308]) returns true; fitWeights with four such choices returns five NaNs. Since each normalized feature lies in [0,1], comparison components must be in [-1,1], or derive the vector server-side from the owned jobs. Reject before persistence and cover repeated large finite vectors.
-
The new lib/profile.ts numberIsAboutEmployer exemption treats any nearby preceding "your" as proof the applicant quantity belongs to the employer. Both "Your role fits my 12 years of experience." and "For your team, I bring 15 years of experience." return isClaim=false and validateDraftLog({body,cited:[]},[],now) accepts them at high confidence. This newly bypasses the unsupported applicant-claim gate. Keep employer-year handling narrow and preserve applicant quantity validation even when employer language appears in the same sentence; add regressions for both mixed-subject cases.
I am separately checking the optimistic-concurrency outcome fix and all UI callers. This review does not claim those checks complete yet. No live records were used in these exported-function reproductions.
|
Additional verification on 8466042: the outcome concurrency blocker is cleared by an independent actual-route probe using SQLite with a D1-shaped adapter. A competing edit injected after SELECT and before the batch returns 409, preserves the competing row, and writes zero outcomes/events. The no-race case returns 200 with exactly one outcome/event. Offer-to-Accepted now validates, and the production Track record caller supplies job.version. The two remaining findings in review5123882135 still require fixes; this is not approval. |
|
One additional user-flow finding confirmed by independent exact-tree search on 8466042: there is no user-accessible submission recording action. app/api/outcomes/route.ts:34-35 excludes Ready jobs from prep; app/track/page.tsx:39-48 offers only later outcomes, and its record call supplies no receipt. No other app/integration caller of /api/outcomes exists. Therefore a newly accepted Ready job cannot follow the page instruction to record a submission with its receipt through the app, and cannot enter the outcome flow. Provide an explicit Ready-job receipt recording action with regression coverage. This records the user's already completed external submission; it should not send applications automatically. |
|
Independent Extra High reproduction on exact All three current blockers still hold. I ran the exported functions from this blob (Node type-stripping, no live records):
Required CI on this head is green and does not clear these blockers. The branch is also behind current |
Reject comparison vectors outside [-1, 1] so oversized finite choices cannot produce NaN weights. Treat applicant quantities as claims even when employer language appears earlier in the same sentence. List Ready jobs on Track with an explicit receipted submission action. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
sdcarlson
left a comment
There was a problem hiding this comment.
Verdict
This PR lands two issues and five new screens (+6137) before anyone has returned to Relay. The actual loops are smaller. Refuse uncited checkable claims. Record that a Ready draft was sent, with a receipt, and that an ended job stays ended. Pairwise logistic regression, E[max] weekly planning, Greenhouse HTML stripping, cluster graduation, and Levenshtein style rules are not required for that. I am not making those cuts a merge condition. The claimed loop is still false on this head until the three holes below close.
Prior 3e081b4 blockers are cleared on 8466042. Offer-to-Accepted, Held-to-Accepted, empty delta:[], and stale outcome 409 all hold in source. CI is green, including calibration and orchestration. I independently reproduced the three remaining holes on this exact tree. No live records.
Merge blockers
isChoiceDeltaaccepts any five finite numbers.[1e308,...]is true.fitWeightsof four such choices returns fiveNaNs, andPOST /api/preferencesstores the client vector. Drop client delta. Recompute it from the two owned job keys. Cover repeated large finite vectors. They must 400 and write nothing.numberIsAboutEmployertreats any nearby precedingyouras proof the quantity is the employer's.Your role fits my 12 years of experience.andFor your team, I bring 15 years of experience.are bothisClaim=false, sovalidateDraftLogaccepts them withcited:[]. Delete the exemption. Every digit in a non-question is a claim. Cover both mixed-subject sentences.- No user-accessible submission recording.
GET /api/outcomesprep excludes Ready./trackonly offers later kinds and never sends a receipt. Workspace still cannot recordsubmitted. The empty copy tells the user to record a submission with its receipt. The control does not exist. Add one Ready-job receipt action. Regression must go through that caller. This records a send the user already made. It must not send.
Do not add more NLP, more fitter steps, or more CI jobs. Those three close the advertised loop.
Ponytail (not merge conditions, do not apply here)
lib/scoring.ts:1 — delete: E[max] planner, beta prior, freshness decay, reach/match/floor labels; a Ready queue under a minute budget is enough.
lib/utility.ts:97 — reframe: stop persisting client delta; winner and loser job keys are the record.
lib/profile.ts:92 — delete: numberIsAboutEmployer and the 56-character window.
lib/profile.ts:189 — yagni: cluster graduation, editRatio, sameness pairs, opening groups, proposeRules; citation refusal plus human review is the product.
lib/postings.ts:1 — yagni: Greenhouse/Lever HTML stripper; existing import already adds evidence.
lib/resume.ts:1 — yagni: resume line parser; paste-a-fact is enough until someone uses it.
app/plan/page.tsx:1 — delete: /plan and /preferences until a human is drowning in Held jobs.
app/agent-tools.ts — shrink: relay_log_draft / relay_read_profile / relay_review_status wait until the three blockers above exist.
.github/workflows/ci.yml — shrink: two new required integration jobs are more surface than #47/#48 need once those tests live in unit.
Unblock
Fix the three blockers on this branch. Re-request review on that head. Do not reopen #26 or merge #46. Do not wait on hosting #3. After this lands, split the ponytail list onto a later issue or delete it.
|
Seth’s latest REQUEST_CHANGES (5123899043) is on superseded Independent Extra High reproduction on exact
Two items from Seth’s newer Unblock still hold on
Green required CI on |
POST /api/preferences records winner and loser job keys and recomputes the comparison from owned posting features. Client vectors, including repeated 1e308 payloads, return 400 and write nothing. Every digit in a non-question is an applicant claim. Ready jobs still record an already completed submission with a receipt. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
sdcarlson
left a comment
There was a problem hiding this comment.
Exact-head follow-up on 179df12: the bounded vector fix passes the previous overflow reproduction (oversized input rejected; fitter stays finite). The previous two claim examples are now rejected, but the same unsupported-claim defect remains under ordinary word order: "I bring your team 15 years of experience." and "I offer your company 12 years of experience." both return isClaim=false and validateDraftLog({body,cited:[]},[],now) accepts at high confidence. Reproduced against this exact source. A nearest-pronoun rule does not establish who owns the quantity. Remove the employer-number exemption rather than expanding the parser again; conservative rejection is preferable to silently admitting unsupported applicant experience. Keep this thread open until these regressions also reject. Track receipt UI is under independent verification.
|
Verified at179df120c55686db0c242f46b849a963717d7407: receipt UI gap is addressed. Independent actual-GET probe returns owned Ready/version7 and excludes Held; Track now offers a receipt field and Record submission passing id/version/kind/receipt, disabling short receipts and busy requests. Existing version-checked POST is unchanged from the independently race-tested fix. Orchestration covers Ready listing and receipt roundtrip; browser-specific interaction was not re-executed in this bounded review. Resolving the two addressed receipt-control threads. Unsupported-claim finding5123907359 remains open; no approval. |
|
Overflow thread reconciliation at179df12: the original requirement allowed bounded normalized components or server derivation. isChoiceDelta now requires exactly five finite components in [-1,1], and the preferences route calls it before persistence. Exact-source repeated1e308 reproduction now rejects and produces finite zero fallback weights; the added selection regression covers this. This satisfies the reported overflow defect, so I am resolving that thread. Server derivation is not an additional merge condition absent a separate concrete defect. The unsupported applicant-claim finding remains blocking. |
Reject uncited experience claims even when employer language sits between the first-person subject and the number. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
|
Seth’s REQUEST_CHANGES 5123907359 is on exact Current head is now
Last-push partner review of |
|
Head moved after Extra High’s Independent Extra High reproduction on exact Seth’s receipt-UI and overflow thread resolutions remain on superseded |
The production job on 91e7e57 failed at owner-a save with the known local-worker restart body. Application suites were already green on this tree; this commit only requeues required checks. Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
|
Required CI on current head |
|
Head is now empty retrigger |
|
Required CI on empty retrigger |
sdcarlson
left a comment
There was a problem hiding this comment.
Approve exact e5c00b3. All reported correctness blockers are addressed. The employer-number exemption is removed and all four mixed-subject examples reject uncited claims. Preference vectors are now derived from owned jobs, both production callers match, and client vectors are rejected before insertion. Independent actual-route probes confirmed both choice directions and no-write rejection. Earlier independent outcome race probe verified post-read conflict returns409 with zero outcome/event inserts; Offer acceptance, editor transition protection and Ready receipt action are fixed. Required calibration/orchestration suites now run on isolated fixtures.
Validation on this exact head: local full unit suite172passed/2platformskips, targeted profile/selection60passed, TypeScript passed. All11 hosted checks are successful, including API/browser/calibration/orchestration/production and CodeQL. The final six-file delta received independent bounded review with no remaining concrete integration finding. This approval covers this reviewed PR, not production deployment or a claim that intermittent #22 is fixed.
<!-- CURSOR_AGENT_PR_BODY_BEGIN --> ## Problem and result #56 reopened `feat/selection-and-outcomes` against current `main`. That branch still contains the already-landed #26/#46/#49 application (profile, plan, track, migrations) plus two newer Claude commits. GitHub therefore reports conflicts on every hot file and does not give a clean “new work” signal. Unique work after the old #46 head (`0d3a006`): 1. **Relay CLI, phase 1** (`login`, `brief`, `log`, `draft`, `plan`, `status`, `outcome`) — this PR. 2. **Employer-number citation exemption** — **not** included. Current `main` already requires a citation for every digit, including `I read your 2024 post on storage engines.` Seth asked that exemption deleted on #49; restoring it would reopen the mixed-subject hole. After this lands, an agent on `pnpm dev` can log drafts through the same citation gate as the browser API. Exit 3 means fix the input and write nothing. Exit 4 means retry. The CLI still cannot accept a draft, verify a fact, close a review, answer a preference pair, or submit an application. Terminal outcomes need `--yes`. Outcomes send the current job `version` so stale records 409 instead of writing. `draft --out` writes only after the citation gate and a successful log. An existing file is left untouched unless `--force` is passed. HTTP 401 is classified as auth (exit 2) before the body is parsed. Non-JSON error bodies use `exitFor(status)`: plaintext 400/403/404/409/413 are exit 3. A malformed 200 stays exit 4. Do not merge #56. Close it as a conflicted duplicate once this replacement is the review target. ## Evidence - On `origin/main` `1b6845b` (#55). Bounded origin drain and #49/#54 behavior are unchanged. - Seth REQUEST_CHANGES on `6bcdcfd`: native `Response('Forbidden', { status: 403 })` threw code 4. Fixed on `71ef3ca` by applying `exitFor` when error-body parse fails. - Local on `71ef3ca`: `pnpm lint` clean; `pnpm typecheck` clean; `pnpm test` **243/243**; `pnpm public-copy:check` current. CLI unit tests 17/17 including native plaintext 400/403/404/409/413 and malformed 200. - Hosted CI on `71ef3ca`: all 12 checks green, including **CI / Integration (cli)** and required **CI / CI**. Run: https://github.com/SyberLabs/relay/actions/runs/34015612819 - Security, data migration, and rollback impact: no schema change. Local-only (`RELAY_URL` must be localhost). Session file is ignored `private-data/.session` mode 600. Production identity is not supported. ## Review - [x] One issue owns this scope; unrelated changes are separate. This is the CLI transport only. - [x] Tests exercise the changed behavior, including failure paths. - [x] No credentials, personal applications, or private research are included. - [x] Agent findings from `2400761` and `6bcdcfd` are fixed. CodeQL threads are left for Seth to resolve with his stated rationale; Extra High does not dismiss them as a security-audit or production-auth approval. - [ ] A maintainer other than the author will approve the **latest push**. Extra High cannot self-approve. - [x] Public capabilities, integration limits and lead-engineer credit match `docs/public-copy.json`; run `pnpm public-copy:check` and follow `docs/PUBLIC-COPY.md` when these change. Do not restore unused UI-kit deps from #56. Do not wait on hosting #3. #22 stays a separate Wrangler flake. <!-- CURSOR_AGENT_PR_BODY_END --> <div><a href="https://cursor.com/agents/bc-3877d527-b5e4-46d7-8c5f-573418309ece?cursor_ref=pr_footer&cursor_cta=open_in_web"><picture><source media="(prefers-color-scheme: dark)" srcset="https://cursor.com/assets/images/open-in-web-dark.png"><source media="(prefers-color-scheme: light)" srcset="https://cursor.com/assets/images/open-in-web-light.png"><img alt="Open in Web" width="114" height="28" src="https://cursor.com/assets/images/open-in-web-dark.png"></picture></a> <a href="https://cursor.com/background-agent?bcId=bc-3877d527-b5e4-46d7-8c5f-573418309ece&cursor_ref=pr_footer&cursor_cta=open_in_cursor"><picture><source media="(prefers-color-scheme: dark)" srcset="https://cursor.com/assets/images/open-in-cursor-dark.png"><source media="(prefers-color-scheme: light)" srcset="https://cursor.com/assets/images/open-in-cursor-light.png"><img alt="Open in Cursor" width="131" height="28" src="https://cursor.com/assets/images/open-in-cursor-dark.png"></picture></a> </div> --------- Co-authored-by: Cursor Agent <cursoragent@cursor.com> Co-authored-by: sykosyber <sykosyber@users.noreply.github.com>
Problem and result
Closes #47
Closes #48
The cleanup queue that deferred closed PR #26 has landed. Open PR #46 already contains #26, but it cannot merge: it conflicts with current
mainonREADME.md,app/globals.css,integrations/connectors.mjs, andpackage.json, and it still carries a stale toolchain.This PR reconciles that work onto current
main(ad4bb8f, including #44, #45, and #51). After merge, a signed-in applicant can:Who needed this: people using assistants at volume, and anyone whose Live loop currently has no honest terminal/receipt model. What failed before: #26 was closed as deferred-not-rejected; #46 was blocked on conflicts. What they can do afterward: use
/profile,/review,/preferences,/plan, and/trackon currentmainwithout dropping session expiry, dirty-editor navigation, null-URL packet identity, recruiter-facing copy, dual-builder credit, or public-copy generation.Supersedes reopening #26. Do not merge conflicted #46 directly. PR #46 commits can stay until this replacement is accepted.
Evidence
3e081b4blockers remain cleared: Offer-to-Accepted, Held-to-Accepted empty draft, emptydelta:[]400, stale outcome 409 with versioned Track caller.49ff152(ponytail list not applied):POST /api/preferencesno longer persists a clientdelta. Winner and loser job keys are the record; the comparison is recomputed from owned Held posting features.delta: []and four repeated[1e308,…]payloads return 400 and leaveansweredunchanged. The preferences UI sends only those two keys.numberIsAboutEmployeris deleted. Every digit in a non-question is a claim. Mixed-subject sentences andI read your 2024 post on storage engines.refusevalidateDraftLogwithcited: [].kind: 'submitted',version,receipt) throughapp/track/page.tsx. Playwright covers that caller. Seth independently verified the receipt UI on179df12and resolved those threads. Overflow on179df12was also resolved as bounded[-1,1]components.179df12: nearest-subject still missedI bring your team 15 years of experience.andI offer your company 12 years of experience.Those sentences reject on49ff152(exemption deleted).91e7e57adds those two strings as regressions. The parser was not expanded.91e7e57failed atowner-a savewith Wranglerworker restarted mid-request(503, process still running). Same tree passespnpm test:productionlocally.e5c00b3retriggers required CI only; Diagnose intermittent 503 in compiled production integration test #22 stays a separate investigation.e5c00b3production re-run):pnpm lintclean;pnpm typecheckclean;pnpm test174/174;pnpm public-copy:checkcurrent;pnpm buildsucceeded;pnpm test:apipassed;pnpm test:calibrationpassed;pnpm test:orchestrationpassed;pnpm test:e2e8/8;pnpm test:productionpassed including migrations0000–0004.Review
One issue owns this scope; unrelated changes are separate.
Tests exercise the changed behavior, including failure paths.
No credentials, personal applications, or private research are included.
Agent findings are fixed or answered with evidence.
A maintainer other than the author will approve.
Public capabilities, integration limits and lead-engineer credit match
docs/public-copy.json; runpnpm public-copy:checkand followdocs/PUBLIC-COPY.mdwhen these change.Landing notes for review
origin/main(ad4bb8f) and resolved public copy in favor of Sharpen recruiter-facing Relay copy and credit both builders #45 wording and credit, plus this PR's capabilities.package.json(Apache-2.0, React 19.2.8, Vinext beta.9, Node 24). Did not restore unused UI-kit deps from Feat/selection and outcomes #46./profile,/review,/preferences,/plan,/track.tests/api.test.mjs,tests/calibration.test.mjs, andtests/orchestration.test.mjseach run on a dedicated freshly migrated D1 in required CI.Standing instruction for agents: #47 and #48 must not be deferred behind hosting #3 or experiment #5.