Repository navigation
chore(repo): remove verified junk from the 2026-10-04 audit - #439
Merged
Merged
Conversation
The audit's input stays untracked; this commit is the dispositions and the deletions. 73 changes whose only tracked references were their own definition or a comment that contradicted the code, across six areas: compiler-proven unused declarations and now-unread parameters in packages/core, two dead playtest/asset helpers plus a dead bench script, the never-packaged Android sample JS and the orphan Metal shader (plus three module-only `export` keywords), twenty zero-reader template helpers, nine example leftovers, and the docs/scripts surface (two unreferenced PDFs, a duplicate audit, two unreachable analysis scripts, five dead exports, two stale round-ledger pointers). Restored on review: the `QuarryScene` class-expression name in examples/quarry. `Constructor.name` is what scene diagnostics print, so the name is diagnostic surface and dropping one identifier cleaned up nothing. Kept, with the reason recorded in the PRD rather than a deletion: the `stamp:template-loading` entry point (its wrapped function has no other caller, so it is the only way to apply a canonical `loading.ts` change to the tracked templates — the role its contract-tested sibling `stamp:template-render.ts` plays), `core/__tests__/seed.spec.ts` (the only proof that a *different* seed diverges), the two playtest compat facades, the published `assets` barrel re-exports, and every owner-decision item. Gates, each with its real exit code: - `pnpm typecheck` exit 0. `pnpm lint` exit 0, 1074 warnings. - `packages/core` `tsc --noEmit --noUnusedLocals --noUnusedParameters` exit 0, empty output (baseline of 24 is the audit input's own measurement). - `pnpm --filter @threenative/runtime-native test` exit 0 — 134 files / 1548 tests passed, 62 skipped; Rust physics parity 21 + 2; publint clean. It needs a compiled host first (`pnpm native:build` exit 0); every pre-build failure was an `is not built` assertion on a missing executable, and none named a file this change deletes. - The four scaffolder specs the restamp touches: 4 files / 93 tests, exit 0. - `pnpm check:docs` exit 0 — 2324 links across 1217 Markdown files. `scripts/__tests__` 143 files / 1944 tests, exit 0. - Generated-surface drift: `generate-ctx-surface-table --check`, `generate-capability-reference --check` and `check-capability-docs --census` all exit 0, so no capability manifest or reference moves. - `pnpm --filter abyss-framework build` exit 0 and `git check-ignore` reports `.gitignore:8:*.build-report.json`, so the build report no longer dirties the tree; `git status --porcelain examples/abyss-framework` after that build lists only this commit's deletions. `pnpm test` is red and its PRD box stays open: the package-test phase is wholly green, and the unit phase is 7932 passed / 13 skipped / 5 failed, all five `Test timed out` in `template-assets-compile-1`, `template-assets-compile-2` and `template.spec.ts`. Those three specs each scaffold, install and build every template, and this box is shared with other agents' worktrees — load average 55–70 on 24 cores during the run. The same four files are green run alone (4 files / 93 tests, exit 0, 205s), so nothing here is a behavioural failure; the PRD records the exact cause and stays in `active/` until one clean `pnpm test` is recorded.
… wall-clock The literal `pnpm test` full board ran to completion with the shipped serial package walk and a four-core pin (1424s): docs, build and package-test green, unit phase 4 failed / 7933 passed / 13 skipped. All four failures are `Test timed out` — the three heaviest scaffolder specs (60s, 60s, 180s) and one native desktop screenshot wait that captured no host output — on a box at load 27-44 across 24 cores. No behavioural assertion failed, and the same four files alone on the four idlest cores are green: 4 files / 117 tests, exit 0, 227s. That is the third run to fail on wall-clock alone, so it is not rolled a fourth time: at launch no core was below 59% busy, and another 24-minute run would only add contention to the box it is trying to measure. The PRD says so, names the pacing that was tried, and leaves AC-2 and the phase-2 gate box unticked because the gate as written ran red. No tracked runner, config, timeout or test changed; `TN_SUITE_PACKAGE_CONCURRENCY` and `taskset` are execution-time inputs the suite already reads. The audit input `junk-audit-2026-10-04.md` stays untracked. PRD stays in `docs/PRDs/tech-debt-code-quality/` — the remaining work is re-runnable, not blocked-only, so it is not a `done/` or `BLOCKED/` move.
…eep moved The normaliseToMetres and attachToBone cells in the PRD-289 applicability table still named Enemy.ts:863 and :869, which this lane's dead-comment deletion moved 15 lines up to :848 and :854. TN_TEMPLATE_CONVENTIONS_FAILED on the budgets job; it is green now. The shooter scaffold hash measured through createProject already matched its pin, so it stays as it is.
…, drop the "only one" claim
Plain merge, no rebase: the merge queue needs a current base. #416 is one commit, 1,087 files; two of them carried cleanup edits too, so one conflict. Resolution of packages/create-threenative/__tests__/scaffold.spec.ts: PRD_201_PARENT_SCAFFOLD_HASHES is re-measured on the merged tree through createProject, not taken from either side. Ten templates moved (the ten this sweep edited, which #416 also touched); platformer, snow and starter keep #416's values because neither side changed them. scaffold.spec.ts 66 tests exit 0. templates/puzzle/src/scenes/Play.ts auto-merged clean and was audited by diff: this lane's `unwatch` deletion and #416's material-lighting wiring are both present. All 14 of this lane's whole-file deletions are gone from the merged tree and every file #416 added is present. Also here: the native census table regenerated by `pnpm census` (android/ 3,187 -> 1,814 is the deleted sample JS and orphan Metal shader), and the PRD-495 record of the merge. `pnpm budgets` needed a rebuilt packages/playtest/dist first — #416's collectRegionalTone was missing from a stale artefact, not from merged source. Verified on the merged tree: pnpm typecheck exit 0; pnpm lint exit 0 (0 errors); pnpm sync:agents --check exit 0; check-template-conventions green; pnpm check:docs exit 0 (2,393 links); scripts lane 8 files / 260 tests exit 0; pnpm budgets exit 0. AC-2 stays unticked — no fresh full board over the merged tree yet. The Windows desktop core leg stays red and is not waived.
jonit-dev
marked this pull request as ready for review
October 5, 2026 01:34
…ere not junk An independent `claude-opus-5-5 --effort high` review checked every removal for readers and found three whose "dead" premise was false, so they come back byte-identical to origin/develop (64aed30): - packages/physics/scripts/bench-allocations.ts — open PRD-191 names it as its Phase 3 EDIT target and verification step 2. Its "NOT STARTED" status is stale: 91e46d3 did that work and its record shows the bench at exit 0. - templates/rain/tools/verify-noise-volume.mjs — the owner's manual storm-volume proof. Nothing in the repo runs it because it needs an external tempest.html; PRD-473 and the release-readiness note cite it as history. - plans/technical-debt-audit-2026-08-20.md — a distinct audit of d75f464, not a duplicate, with P0-2 and P0-4 still tracked by PRD-177/184 and PRD-180. The rain hash is re-measured, not guessed: the byte-stability assertion printed the generated tree's actual value. Executed changes drop 73 -> 70 (B 5->4, D 20->19, E 11->10) and each restoration is now a Retained row with its reason. Also corrects three claims the review disproved: metal_rt.mm embeds only raytracingKernel, VISIBILITY_INTENT is still read in rain/src/game.ts, and the hash comments no longer claim one declaration per template. The Windows bullet now names the warmup advance(60) before the tolerant startup wait as the fault and withdraws the self-imposed merge gate — that leg needs native_tier 'full' and a develop pull request classifies 'reduced' (scripts/ci-change-scope.mjs:610-618), verified against this pull request. The broader full-platform audit stays recorded as red, and AC-2 stays unticked.
…D as done CI run 37255318117 at head 346afd0 reached terminal success with typecheck, lint, test, test-native, all four test-unit shards and ci-required green, so AC-2 and its Phase 2 workspace-gate box now have their proof. The literal local `pnpm test` never went green and is still recorded as red on wall-clock alone; the tick rests on the hosted run, which covers the same four run-test-suite phases over the same package scope. The Windows desktop-core leg stays red history and skipped here under the reduced native_tier, not waived. Filed to docs/PRDs/done/; the PRD-208 link is re-pointed.
jonit-dev
enabled auto-merge
October 5, 2026 03:28
jonit-dev
disabled auto-merge
October 5, 2026 04:08
jonit-dev
enabled auto-merge
October 5, 2026 04:16
jonit-dev
disabled auto-merge
October 5, 2026 04:24
9 tasks done
jonit-dev
added a commit
that referenced
this pull request
Oct 8, 2026
Origin filed PRD-495 for the 2026-10-04 junk audit (#439) while the local branch gave 495 to the compile-walk PRD. The compile walk takes 540, the next free number, and every link follows.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Removes verified unused engine declarations, unreachable template/example source, uncompiled native samples, and unreferenced repository artifacts from the 2026-10-04 junk audit. Every finding has a recorded disposition; 70 cleanup changes remain. Public exports, live compatibility surfaces, diagnostic names, and owner-decision items stay intact.
Opus 5.5 at high effort reviewed the complete deletion diff. The physics allocation benchmark still supports PRD-191 and is restored. The rain noise verifier and distinct August audit are retained conservatively for their proof/history value. All three match develop byte-for-byte; the restored benchmark ran successfully and the rain scaffold hash was measured again.
PRD-495 is complete: 7/7 phase boxes and 6/6 acceptance criteria. The last two commits only close/archive that record; source code is unchanged from the qualified head.
Validation:
The broader manual audit exposed a pre-existing Windows warmup/mailbox defect. Repository policy selects reduced native coverage for develop PRs and their merge queue; all selected Linux/Android and complete unit/package gates remain required. Windows/macOS desktop core and iOS are skipped under that policy. The broader audit failure stays recorded.
The final documentation head must pass its fresh checks and the required merge queue before landing.