@@ -227,118 +227,7 @@ const SKIP_VERSIONS = new Set(['3.8.0']);
227227 * latest, when `previous.version` matches via the baseline fallback). Once
228228 * a version is no longer the latest in committed history and no longer the
229229 * baseline used for `dev` comparisons, its entries become dead weight and
230- * should be removed (last pruned: 3.9.0/3.9.1/3.9.2/3.9.6/3.10.0).
231- *
232- * - 3.11.0:Query time / 3.11.0:No-op rebuild / 3.11.0:fnDeps depth 3 /
233- * 3.11.0:fnDeps depth 5 — CI runner variance on sub-50ms WASM metrics
234- * when the per-PR gate replays dev against the just-published 3.11.0
235- * baseline. The dev source tree between commit f7c29c5 (3.11.0 release)
236- * and the post-publish docs PR (#1217) contains only version-bump diffs
237- * in package.json/Cargo.toml/package-lock.json — no extractor, query,
238- * or DB changes. Published 3.11.0 numbers were captured by the publish
239- * workflow on one runner; the per-PR gate re-measures on a fresh runner
240- * and lands ~10ms higher on every sub-50ms WASM metric:
241- * - Query time: 32.5 → 44.2 (+36%) on run 26426483639
242- * - No-op rebuild: 18 → 29 (+61%)
243- * - fnDeps depth 3: 33.6 → 44.1 (+31%)
244- * - fnDeps depth 5: 33.3 → 44.4 (+33%)
245- * The 3.11.0 vs 3.10.0 release-time comparison shows these metrics
246- * either flat or improved (Query time 37.6 → 32.5 = -14%, fnDeps depth 3
247- * 33 → 33.6 = ~flat, fnDeps depth 5 33 → 33.3 = ~flat, No-op rebuild
248- * 15 → 18 = +20% but at runner noise floor) — confirming no underlying
249- * regression and ruling out a real slowdown introduced in 3.11.0. Same
250- * shape and root cause as the pruned 3.10.0 entries. Exempt this release;
251- * remove once 3.13.0+ data confirms the new steady-state on whatever
252- * runner generation is current at that time.
253- *
254- * - 3.11.0:1-file rebuild — CI runner variance on a sub-100ms native
255- * incremental metric. The 3.11.0 baseline was captured at 64ms; the
256- * per-PR gate re-measures dev on a fresh runner and can land +45ms
257- * higher (e.g. 64 → 109ms = +70%, threshold 50%) on runs where the
258- * runner is under load. No incremental, Rust, or JS change in the
259- * dev tree accounts for this delta — other dep-only PRs running
260- * concurrently measured 64 → 80ms (+25%) on the same corpus
261- * (run 26706695868), confirming per-runner noise rather than a
262- * structural slowdown. The 50% NOISY_METRIC_THRESHOLD is razor-thin
263- * for a sub-100ms metric at shared-runner noise floor (~20ms). Exempt
264- * this release; remove once 3.13.0+ data confirms stabilization.
265- *
266- * - 3.11.0:Full build — same CI runner-variance root cause as the four
267- * 3.11.0 entries above, but on the multi-second WASM full-build metric
268- * rather than the sub-50ms group. Surfaced on the embedding-bench docs
269- * PR (#1218) when the per-PR gate re-measured dev (byte-identical to
270- * released 3.11.0 modulo EMBEDDING-BENCHMARKS.md) against the published
271- * 3.11.0 baseline:
272- * - Full build (wasm): 7664 → 9765 (+27%, threshold 25%) on run 26431397916
273- * Historical WASM full-build numbers on the same corpus span 7.2s–14.0s
274- * across 3.9.0–3.11.0, so 9.8s on a single dev re-measurement sits well
275- * inside the runner-noise envelope (3.10.0 baseline was 8.4s, 3.9.6 was
276- * 14.0s). PR #1217's gate measured 7664 — the same code on a fresh
277- * runner instance later measured 9765. No extractor, builder, or DB
278- * layer changed between 3.11.0 release and #1218; only EMBEDDING-
279- * BENCHMARKS.md, which is not loaded at build time. Exempt this release;
280- * remove once 3.13.0+ data confirms stabilization under whatever runner
281- * generation is current at that time.
282- *
283- * - 3.11.1:DB bytes/file — same methodology-scope artifact as 3.10.0:DB
284- * bytes/file. The 3.11.0 release does not have query benchmark data
285- * committed to history, so findLatestPair falls back to 3.10.0 as the
286- * baseline. The 3.10.0 corpus included resolution fixtures (~745 files);
287- * 3.11.1 measures only the codegraph source (~607 files) after #1134
288- * excluded resolution fixtures from the build sweep. The denominator
289- * shrinks while total DB content stays roughly constant, inflating
290- * dbSizeBytes/file: native 41614 → 54107 (+30%), wasm 41543 → 53517
291- * (+29%). No schema or extraction change; remove once 3.13.0+ data is
292- * captured with the full 3.11.x baseline in committed query history.
293- *
294- * - 3.11.1:fnDeps depth 3 / 3.11.1:fnDeps depth 5 — same baseline-gap
295- * root cause as 3.11.1:DB bytes/file. Because 3.11.0 query benchmark
296- * data is absent from committed history, the guard compares 3.11.1
297- * against the pre-3.11.0 3.10.0 baseline. The 3.10.0 query numbers
298- * predate the steady-state established in 3.11.0 (fnDeps depth 3: 33ms,
299- * depth 5: 33ms), so 3.11.1's equivalent values appear as regressions:
300- * - native fnDeps depth 3: 24.3 → 34.7 (+43%)
301- * - native fnDeps depth 5: 24.7 → 34.7 (+40%)
302- * - wasm fnDeps depth 3: 33 → 43.2 (+31%)
303- * - wasm fnDeps depth 5: 33 → 43.5 (+32%)
304- * No fn_deps Rust implementation, fnDepsData JS wrapper, or DB index
305- * changed between 3.10.0 and 3.11.1. Remove once 3.12.0+ data confirms
306- * stable query numbers against a 3.11.x baseline.
307- *
308- * - 3.11.2:No-op rebuild — CI runner variance on a sub-30ms native metric.
309- * The 3.11.2 baseline captures noopRebuildMs=25 (build benchmark) and
310- * noopRebuildMs=19 (incremental benchmark); the per-PR gate re-measures
311- * dev on a fresh runner and lands at 45ms (+80%) and 37ms (+95%) on run
312- * 26792023287 — both exceed the NOISY_METRIC_THRESHOLD of 50% due to
313- * sub-30ms variance. No watcher, builder, or incremental-orchestrator
314- * change is present in the dev tree for this docs-only PR (#1282);
315- * the delta is entirely shared-runner scheduling noise. Same shape and
316- * root cause as the 3.11.0 and 3.11.1 entries above. Exempt this
317- * release; remove once 3.13.0+ data confirms the steady-state.
318- *
319- * - 3.11.2:1-file rebuild — CI runner variance on the sub-100ms native
320- * incremental metric. The 3.11.2 baseline was captured at 83ms; the
321- * per-PR gate for the Phase 8.1 TypeScript resolver PR (#1278) re-measured
322- * dev on a fresh runner and landed at 212ms (+155%, threshold 50%) on run
323- * 26793082961. The same PR modifies only: (a) a new ts-resolver.ts module
324- * gated behind `typescriptResolver: false` (was the default at the time), (b) an import of
325- * that module in build-edges.ts, and (c) a config field — none of which
326- * execute on the incremental hot path. Locally the same PR measures 86ms
327- * (within noise of the 83ms baseline). The 3.11.0:1-file rebuild exemption
328- * above documents the same pattern for the same baseline range. Exempt
329- * this release; remove once 3.13.0+ data confirms stabilization.
330- *
331- * - 3.11.2:Full build — CI runner variance on the multi-second native
332- * full-build metric. The 3.11.2 baseline captures fullBuildMs=2231; the
333- * per-PR gate for the Phase 8.3c parameter-flow PR (#1308) re-measured dev
334- * on a fresh runner and landed at 2852ms (+28%, threshold 25%) on run
335- * 27003863932. The PR adds CHA post-pass and parameter-flow tracking that
336- * each contribute microseconds-level overhead per call site — not hundreds
337- * of milliseconds. Historical native full-build numbers on this corpus span
338- * 1959ms (3.10.0) to 2986ms (3.9.6), so 2852ms sits well within the
339- * runner-noise envelope. Same shape and root cause as the 3.11.0:Full build
340- * exemption above (which was a WASM metric; this is native). Exempt this
341- * release; remove once 3.13.0+ data confirms the steady-state.
230+ * should be removed (last pruned: 3.9.0/3.9.1/3.9.2/3.9.6/3.10.0/3.11.0/3.11.1/3.11.2).
342231 *
343232 * - 3.12.0:No-op rebuild — CI runner variance on a sub-50ms native metric.
344233 * The 3.12.0 baseline captures noopRebuildMs=30 (build benchmark) and
@@ -404,18 +293,6 @@ const SKIP_VERSIONS = new Set(['3.8.0']);
404293 * (DB bytes/file), neither of which the WASM widening covers.
405294 */
406295const KNOWN_REGRESSIONS = new Set ( [
407- '3.11.0:Query time' ,
408- '3.11.0:No-op rebuild' ,
409- '3.11.0:1-file rebuild' ,
410- '3.11.0:fnDeps depth 3' ,
411- '3.11.0:fnDeps depth 5' ,
412- '3.11.0:Full build' ,
413- '3.11.1:DB bytes/file' ,
414- '3.11.1:fnDeps depth 3' ,
415- '3.11.1:fnDeps depth 5' ,
416- '3.11.2:No-op rebuild' ,
417- '3.11.2:1-file rebuild' ,
418- '3.11.2:Full build' ,
419296 '3.12.0:No-op rebuild' ,
420297 '3.12.0:Full build' ,
421298 '3.12.0:1-file rebuild' ,
0 commit comments