diff --git a/.codex/plans/2026-10-13-public-companion-closeout.md b/.codex/plans/2026-10-13-public-companion-closeout.md deleted file mode 100644 index 3561685..0000000 --- a/.codex/plans/2026-10-13-public-companion-closeout.md +++ /dev/null @@ -1,22 +0,0 @@ -# Public companion and dependency closeout - -Land the existing public companion PR 61 on main, preserving its PR 59 dependency -and the historical edition. Coordination PRs 5 and 8 remain separate. - -1. Reproduce still-valid PR 59 findings with pure and real-browser regressions. - Preserve interrupted exposure, first-visit geometry, retained study conditions - and observations, read-only diagnostics, suspended clocks and context traces. -2. Correct the reviewed implementation and inaccurate evidence descriptions. - Keep all existing acceptance checks; use hosted browser verification while - another admitted workload owns the local heavy slot. Do not disturb that job. -3. Publish corrections without rewriting history, integrate them into the stacked - PR 61, validate its functional API and contract, and submit exact heads once - through the governed merge process in dependency order. -4. Preserve obsolete authored tips through verified remote/local archive tags, - clear spent refs only after recoverability checks, synchronize local main and - every retained branch/tag, and publish measured results in the existing PRs. - Retain this code-only checkout within the two-checkout 256 MiB allowance. - -Historical media is excluded from the sparse checkout. Original media, artwork -publication, physical-device acceptance and unrelated review lanes remain owned -by their existing records. No source retirement or history rewrite is included. diff --git a/.gitignore b/.gitignore index 58f33d6..abc5216 100644 --- a/.gitignore +++ b/.gitignore @@ -6,3 +6,13 @@ playwright-report/ test-results/ verification/rerun/ *.bundle + +# Agent work records are kept in a private archive, not in public repos. +.claude/plans/ +.claude/handoffs/ +.codex/plans/ +.codex/receipts/ +.gemini/plans/ +.gemini/handoffs/ +.conductor/ +.specstory/ diff --git a/experimental/six-encounters/docs/REVIEW_CLOSEOUT.md b/experimental/six-encounters/docs/REVIEW_CLOSEOUT.md deleted file mode 100644 index 1001f29..0000000 --- a/experimental/six-encounters/docs/REVIEW_CLOSEOUT.md +++ /dev/null @@ -1,43 +0,0 @@ -# Review reconciliation, 13 October 2026 - -The dependency's outstanding findings were checked against its final renderer. -The pure interrupted-exposure regression failed with `0 != 1234` before the fix. -All eight actual-browser regressions failed against head `553c2c9` in hosted run -38460536541 before runtime changes. Local browser admission was denied because -another live job owned the heavy slot; that job was preserved. Hosted checks use -the existing pinned Playwright dependency and keep every prior acceptance gate. - -Corrections preserve interrupted exposure without inventing a return; capture -first-visit geometry before immediate Return; retain study conditions, records -and observations across index navigation; update visible serialization on input; -clone diagnostic records; avoid a new trial when leaving Reading view; suspend -frame scheduling while a clock is ineligible and resume through existing events; -and select entrance traces from the displayed score/generative context. - -The baseline geometry test now independently checks column times cell width. -The memory test reaches a genuine second visit before attempting an overwrite. -The color specification uses the delivered palette. The plan distinguishes the -recorded private preview deployment from production publication. Both the -character test's emitted claim and its historical report now describe exactly -what the early-departure assertion proves: no generation remains active. - -The earlier readiness race, unused motion-family and old screenshot findings -were already addressed by the final reviewed renderer/evidence before this -follow-up; its phase guard, family-specific animation and final captures remain. -The static-analysis HTML findings concern bounded identifiers, numbers and text -from immutable shipped content/configuration, not external or reader input. -Reader observations are serialized through `textContent` and never interpolated -into HTML. Those duplicate alerts do not establish a reader-controlled HTML sink. - -Passing software checks do not establish artist, physical-device or release -acceptance. PRs #59, #61, #62, #63, and #64 are merged into `main`. - -## Preview Recovery & Rollback Rehearsal (Issue #20) - -Deterministic static verification and rollback checks confirm clean deployment recovery: -1. **Static Surface Determinism:** Verify that all HTML, script modules, and style sheets are self-contained without remote runtime dependencies or dynamic network fetches. -2. **Deterministic Test Verification:** Verify test contracts across core engine and generation modules using `node --test tests/generation.test.cjs tests/state.test.cjs`. -3. **Rollback Rehearsal:** - - In case of preview degradation or regressions, the preview deployment can be deterministically rolled back to the prior known-good commit on `main` (e.g. `068ef11` or specific release tags). - - Asset integrity is verified via `source-census.json` checksums, ensuring no mutated or unverified media blobs contaminate the preview environment. - - Preserves state machine immutability: session state and local playback history are stored strictly in session memory and can be cleanly re-initialized on reload.