Add C4 release evidence dashboard - #87
Conversation
Replace narrativeState.hooks z.array(z.any()) with a strict NarrativeHookSchema verified against the NarrativeHook interface and both writers (franchise-week-helpers weekly refresh, convention-save seed). type stays a free-form string: hooks-engine categories are an open set. Adds 5 schema tests.
New scripts/evidence-dashboard.mjs (pnpm evidence:dashboard) refreshes a marked Evidence Dashboard section at the top of STATUS.md: remote CI gate per-check conclusions, G6 route-sweep contract presence, Ecology Lab nightly scoreboard, local gate contract, bundle size vs ceiling, Math.random scan, opt-in playtest anomalies, and ledger freshness. --check exits non-zero on red rows for future CI use.
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 60ddf16d1c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const job = jobs.find((j) => j.name === name); | ||
| return `${name}:${job ? job.conclusion : '?'}`; | ||
| }); | ||
| const status = ci.conclusion === 'success' ? OK : ci.status !== 'completed' ? WARN : FAIL; |
There was a problem hiding this comment.
When the CI workflow succeeds after one of the expected jobs is renamed or removed, this still marks the remote gate ✅ even though the evidence text reports that job as ?. Because this row is meant to attest specifically to test, determinism-gate, and release-gate, derive the status from all three named job conclusions as well as the workflow conclusion; otherwise a weakened CI contract is displayed as healthy.
Useful? React with 👍 / 👎.
| const gateList = tryExec(process.execPath, ['scripts/release-gate.mjs', '--list']); | ||
| const hasSweep = /g6|route sweep/i.test(gateList.out); |
There was a problem hiding this comment.
If the visual route-sweep steps are removed while any other G6 step remains, this broad regex still turns the route-sweep row green. The inspected release plan contains unrelated g6-chip, g6-ux, and g6-mobile steps, so the check should look for the specific g6-visual-sweep-* IDs or g6-visual group before claiming that the visual/playability sweep remains in the contract.
Useful? React with 👍 / 👎.
| evidence: gateList.ok && hasSweep | ||
| ? 'G6 sweep step present in the 37-step release contract (runs inside every release-gate, local + remote)' | ||
| : 'G6 sweep step NOT found in release-gate contract', | ||
| pointer: 'node scripts/release-gate.mjs --only browser (local sweep)', |
There was a problem hiding this comment.
The advertised refresh command does not run the G6 route sweep: inspecting scripts/release-gate.mjs --only browser --list selects only built-page-smoke, g1-full-setup-desktop, and g5-cba-commissioner-browser. The visual sweep is in the g6-visual group, with its mobile counterpart under g6-mobile, so following this dashboard instruction cannot refresh the evidence named by the row.
Useful? React with 👍 / 👎.
| pointer: 'pnpm --filter @mfd/web build && bash scripts/check-bundle-size.sh', | ||
| }); | ||
| } else { | ||
| const chunk = readdirSync(distDir).find((f) => /^engine-.*\.js$/.test(f) && !f.startsWith('engine-content-')); |
There was a problem hiding this comment.
When a build contains multiple non-content engine-*.js chunks, find silently measures whichever filename readdirSync returns first and may report ✅. The authoritative scripts/check-bundle-size.sh explicitly fails unless exactly one matching engine chunk exists, so this dashboard can disagree with the real release gate after an accidental chunk split; count the candidates and fail the row unless there is exactly one.
Useful? React with 👍 / 👎.
2 participants
Understanding
C4 (Fable board #14): "is the ledger current?" was ambiguous — gate results, bundle size, route sweep, and anomaly counts lived in different places with no single freshness view.
What changed
scripts/evidence-dashboard.mjs(alsopnpm evidence:dashboard) refreshes a marker-delimited## Evidence Dashboardsection at the top ofSTATUS.md. Idempotent: reruns replace only the marked section.gh), G6 route-sweep presence in the 37-step release contract, Ecology Lab nightly scoreboard, local release-gate contract integrity, engine-chunk bundle size vs the 312 KB ceiling (when a local build exists), Math.random determinism scan, opt-in fast-tier playtest anomalies (--with-playtest), and run-ledger freshness.--checkexits non-zero on any red row (future CI hook).STATUS.mdgains the dashboard section plus a run-ledger entry (which also flips the freshness row green).package.jsongainsevidence:dashboard.What did not change
Verification
node scripts/evidence-dashboard.mjs10d9da3; idempotent across reruns ✅node scripts/evidence-dashboard.mjs --checkRemote 37-step gate runs on this PR as a required check.
Rollback
Revert this commit; removes the script and STATUS.md section only.