You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: docs/releases/v0.10.0-implementation-ledger.md
+1Lines changed: 1 addition & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -24,6 +24,7 @@ that npm, GitHub `main`, or the public Action already contains the fix.
24
24
|#640 prior FixMap/setup artifacts | implemented | Contract-recognized report/context/verify artifacts and signature-recognized setup commands are removed from the analysis file and changed-file snapshots before reporting. Same-path team-owned content remains eligible. Artifact and scan-to-analysis tests cover both sides. |
25
25
|#641 routed-test attribution | implemented | Impact routing first attributes a test to the ranked seed it actually imports, then uses path proximity only as a fallback. A regression with an earlier unrelated `data.json` seed proves the test remains attributed to its imported source. |
26
26
|#642 Windows npm doctor shims | implemented | Doctor collapses same-directory extensionless/`.cmd`/`.ps1` npm launchers into one installation while retaining identical launchers from different directories as a real conflict. Focused doctor tests pass. |
27
+
| Post-8446838 correctness sweep | implemented | Six claims from an external 100-item audit survived live-branch reproduction and now have focused regressions: diagnostic truncation backs off rather than splitting UTF-16 surrogate pairs; tsconfig aliases preserve an empty wildcard capture; `sk-` secrets ending in `-` redact correctly; cached and parsed history accepts exact 40-character SHA-1 or 64-character SHA-256 identities; explicit single-ref diffs no longer absorb unrelated untracked files (working-tree inclusion remains separately opt-in); and remote Git refs cannot start with `/`. Stale, speculative, and disproven claims from that document were not treated as bugs. |
27
28
| Scheduled external baseline snapshot | implemented | The 2026-08-24 `external-eval` log confirms both FixMap gates passed and only `evaluate-baseline --suite external --check-recorded` failed on `main`. The branch's deterministic-evidence work refreshes that artifact. After the remaining language and scan corrections, a full 16-repository record changed only six lower-ranked webpack evidence scalars; every Top-5 path and aggregate hit rate stayed unchanged, and two subsequent filtered webpack runs were byte-identical. A clean Linux check remains a candidate gate. This repairs snapshot drift, not a general benchmark claim. |
28
29
| Nested-checkout history scope | implemented | A generated-example freshness gate exposed that a scan rooted below a parent Git checkout read the parent-wide commit log and could confuse matching parent-relative filenames with repository-relative identities. History counts and names are now path-limited and `--relative` to the scanned root; a nested-repository regression proves unrelated parent commits are absent. |
29
30
| PR CI and v0.10 stress campaign | verified | Draft PR #645 runs the real GitHub matrix; current checkpoints through run 33295260500 pass the comprehensive `npm run ci` job plus Node 20.11/22 Ubuntu and Node 24 macOS/Windows jobs. Core and CLI Vitest configs bound file-level workers and retain 15-second default integration/hook timeouts. A sustained 94%-CPU Windows run reproduced seven Git/cache tests at their 30-second ceiling; those named heavy cases now allow 60 seconds only on Windows and remain 30 seconds elsewhere, after which the isolated group and the complete 736-test Core suite passed without cache races. The MCP wire stress retains 10 seconds off Windows and allows 30 seconds for Windows process startup. `npm run stress:v0.10` proves four concurrent cold analyses are deterministic, exact warm reuse is faster, corrupt indexes heal without stale source, saved reports stay isolated, outside links are not scanned, and MCP returns `-32002`/`-32700` on the wire. The scanner benchmark now requires both its completion marker and fixture directory before reuse, preventing a stale marker from producing a false zero-file failure. A final uninterrupted local `npm run ci` passes 42 Action, 313 CLI, 736 Core, and 4 web tests plus audit, lint, production build, generated artifacts, smoke, stress, study integrity, retrieval, adversarial, and the 1,000-file scanner bound. Clean-package tarball proof and the broader release-candidate matrix remain separate gates. |
0 commit comments