Repository navigation
chore(ci): refresh complete measured browser shard weights - #2379
Conversation
schlessera
added
type: chore
Complete ready-candidate proof — 2026-10-18Automatic confirmation39076714711, attempt1, succeeds with all16 jobs successful and fresh browser/layout execution. Metadata requests base
Both jobs meet the issue's <=840-second acceptance including setup, discovery, receipts, upload and finalization. Their sampled VM resources use four logical CPUs and16.77GB RAM; sampled peaks11.42/11.58GB and active CPU1,819.43/2,044.63seconds include container work and are not exact process peaks. No deadline or concurrency changes. Downloaded archives independently match API SHA-256: artifact12117825852 Current event costs9,373 summed executed job-seconds /1,176 event-to-last-job seconds, excluding separate draft,contract,project-sync and local work. The earlier same-base unmodified-main cancellation and historical successful measurement source retain distinct trees/events. This is a successful changed-partition capacity observation, not a randomized causal or whole-CI speedup estimate. The narrowest observed margin is63seconds; actual-main repetition is still required and may falsify qualification. Executed jobs and actual checkout receipts:
PR #2379 and issue #2366 retain release0.44.0. PR implementation/candidate proof is qualified for matching-head merge; #2366 remains open under Refs until exact-squash automatic-main proof satisfies complete selected membership and both browser jobs <=840seconds. Dependant blockers remain until that evidence is reconciled. |
Completed delivery and actual-main qualification — 2026-10-18PR #2379 is delivered as Separate automatic main-push 39078216213, attempt1, succeeds with all16 selected jobs successful, including fresh browser/layout execution, complete units, both native runtime categories, endurance, captures, packaging and strict verification. All16 actual checkout logs agree on the exact squash above. Independent aggregate117533501679 verifies every selected membership and succeeds. No PR result or retained execution substitutes for main proof.
Main shard1 measures63.809seconds build /547.632seconds complete pinned suite /482.40seconds Vitest; shard2 measures79.610 /622.728 /549.92seconds. Both actual runner queue waits are3seconds. Whole-job elapsed includes setup, discovery, receipt computation, upload and finalization. Four logical CPUs and16.77GB RAM are recorded; sampled whole-VM peaks12.92/12.00GB and active CPU1,190.28/1,612.68seconds include containers and are not exact process peaks. Both original15-minute job limits, two shards and per-project concurrency remain unchanged. Downloaded main browser archives independently match API SHA-256: artifact12118056529 Main costs 9,321 summed executed job-seconds /1,238 event-to-last-job seconds; ready costs9,373 /1,176. These exclude separate draft,contract,project-sync and local work. Main's browser jobs are faster than ready while overall elapsed is longer, illustrating why the observed capacity qualification is not a whole-CI or randomized causal speedup claim. Historical cancelled identities and the independent #1229 observation retain their original attribution. No unchanged retry, extra shard, timeout increase, assertion reduction or guard/pin weakening was used. Actual main job checkout receipts, all at
All #2366 acceptance criteria are satisfied: validated successful measurement source; browser-only refresh with all56 layout costs/provenance preserved; real complete current discovery and exactly-once partition; original limits/assertions/pins/guards/artifacts; focused92/479 controls, intended mutation/restoration and required956/6,328 preflight; fresh automatic ready/main proof with at least60seconds observed whole-job headroom; exact checkout/case/resource/queue/cost/aggregate reconciliation. No consumer package or contract change required a changeset. PR and issue retain release0.44.0. This completes #2366. Its capacity dependency is satisfied for #1855/#1942/#2118 and their preserved draft implementations. Their own acceptance and fresh candidate qualification remain; #1942 still has its independent #2365 blocker. No dependent PR is promoted or merged by this receipt, and neither parent epic is claimed complete. This invocation's claim is released on closure; original/local evidence is retained outside the source tree. |
1 participant
Refs #2366
What changed
Browser sharding now uses all 309 measured browser project/file costs from successful automatic run39062246171, attempt1. Current discovery changes the split from 231/78 to 138/171 while executing every emitted pair exactly once. Layout's 56 costs and complete provenance are unchanged. Two jobs, the 15-minute outer bound, per-project concurrency, all assertion deadlines, pins, offline/native/Worker guards and diagnostics remain unchanged.
Why this shape
#1788 measured substantial default-weight and historical-weight drift. Its bounded recommendation refreshes the existing table before adding resources or changing budgets. The existing refresh validator admitted only successful workflow39062246171 and two fresh successful browser jobs117482652349/117482652319, actual checkout
e84e683b01b9cd4a01f3195d36e6f01942dce42a, tree6fb1e0e35a2751b15cb897ab3baf0d712cfdba57, parents2a0b77f675da2f01c32adba3f4920650a5ee09fa/d0a65205db16fc05486e6a0cf46ef61650a6ca6f. Cancelled runs remain diagnostic observations rather than measurement sources. Historical replay predicts a shorter serial path; actual runtime improvement remains unproved.Proof
2d94cb95cwas prepared on main498270fbdc4ef2788fdd3babb259b647d77f6c77. The intervening CONTRACT: fix(ui-sdk): qualify separate-UID native maintenance inputs #2361 native-maintenance source was reviewed; its distinct base changes are inherited, not part of this table-only diff.bun scripts/refresh-browser-costs.ts 39062246171 1 browserrevalidates successful workflow/job/checkout/tree/parent, complete report inventory and source-declared immutable image. Generated values/provenance deeply equal the investigated proposal. The actual CLI adds its normal terminal LF: file SHA-256ac6ecbb657d1bca5d97828434661b231f34dbe91463ddb0f3e709531674a750a; the spike'sce538...hash described the same JSON without that LF.cancelled layout job cannot supply admitted measurements despite passing file logsfails because the promise resolves instead of rejecting. Restoring the source passes1test/1assertion.emitted aggregate executable fails a cancelled selected job and accepts its corrected receipt. Restoring passes1test/4assertions. Both mutations were restored before this table-only commit; no module-load error is counted as red proof.2d94cb95cadbc748b7e82a716799a88d4b50e94band merge base498270fbdc4ef2788fdd3babb259b647d77f6c77.82a64a13a08c44f2daef425d1a9df2c10840ea8evia test: keep worker-probed TCP listeners off the ephemeral range #2371: five native TCP-listener test/helper files avoid ephemeral-port collisions. Their full diff was reviewed; no browser/table/workflow/pin/build inputs change and the table applies without overlap. Local receipts retain the actual source above; fresh full hosted confirmation must test the new combined candidate and all affected native controls. No historical local/CI receipt is relabeled as that combined proof.498270fbdc4ef2788fdd3babb259b647d77f6c77, tree7a992ffc27efb2fecd3dacc9f9a27b0b1ee1edfc. Other selected execution jobs pass; the aggregate correctly rejects cancellation. This is pre-correction evidence, not a passing main run.Checklist
Anything left open
Nothing for #2366. Required ready-candidate and exact-squash automatic-main proof are complete, including both fresh browser jobs within840seconds and unchanged case inventories. Other issues retain their own acceptance criteria.
Complete ready-candidate proof — 2026-10-18
Automatic confirmation39076714711, attempt1, succeeds with all16 jobs successful and fresh browser/layout execution. Metadata requests base
82a64a13a08c44f2daef425d1a9df2c10840ea8e. All16 actual checkout logs agree onde0849c667a166ca873d16eeef4801a96ba78ca0, exact immutable parents that base and head2d94cb95cadbc748b7e82a716799a88d4b50e94b, treeb31b2d1b0d75f0eaa409c4bb3a83d8ac5fca2b2d; fetched Git and immutable API evidence agree. The independent aggregate117528944522 passes every selected category. Latest contract run39076668889 also succeeds; earlier draft/superseded cancellations retain their original event scope.Both jobs meet the issue's <=840-second acceptance including setup, discovery, receipts, upload and finalization. Their sampled VM resources use four logical CPUs and16.77GB RAM; sampled peaks11.42/11.58GB and active CPU1,819.43/2,044.63seconds include container work and are not exact process peaks. No deadline or concurrency changes.
Downloaded archives independently match API SHA-256: artifact12117825852
2139f589bae54c6be84e4609b9b7bc2dad516edcd6a0458bc7a46db0a4935bf5, artifact1211713967766d62b86cad48e5c0fca6a4213b1fc3e76b467d5788648ed1dcb7e2825c2769a. Their collected/module/case manifests reconcile exactly309 distinct project/file pairs,9,050 passes,ten existing pointer skips,no unhandled errors. Every original successful measurement source's per-file case ID/name/state set deeply equals the new execution set, despite redistribution. No omission or new skip is hidden by totals.Current event costs9,373 summed executed job-seconds /1,176 event-to-last-job seconds, excluding separate draft,contract,project-sync and local work. The earlier same-base unmodified-main cancellation and historical successful measurement source retain distinct trees/events. This is a successful changed-partition capacity observation, not a randomized causal or whole-CI speedup estimate. The narrowest observed margin is63seconds; actual-main repetition is still required and may falsify qualification.
Executed jobs and actual checkout receipts:
de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.de0849c667a166ca873d16eeef4801a96ba78ca0.PR #2379 and issue #2366 retain release0.44.0. PR implementation/candidate proof is qualified for matching-head merge; #2366 remains open under Refs until exact-squash automatic-main proof satisfies complete selected membership and both browser jobs <=840seconds. Dependant blockers remain until that evidence is reconciled.
Completed delivery and actual-main qualification — 2026-10-18
PR #2379 is delivered as
63fde3c8fe76e754aa4524f95b23f9786f4447fe. Actual squash parent82a64a13a08c44f2daef425d1a9df2c10840ea8eand treeb31b2d1b0d75f0eaa409c4bb3a83d8ac5fca2b2dexactly match the qualified combined candidate. Ancestry on fetched remote main is verified. No merged branch received another push.Separate automatic main-push 39078216213, attempt1, succeeds with all16 selected jobs successful, including fresh browser/layout execution, complete units, both native runtime categories, endurance, captures, packaging and strict verification. All16 actual checkout logs agree on the exact squash above. Independent aggregate117533501679 verifies every selected membership and succeeds. No PR result or retained execution substitutes for main proof.
Main shard1 measures63.809seconds build /547.632seconds complete pinned suite /482.40seconds Vitest; shard2 measures79.610 /622.728 /549.92seconds. Both actual runner queue waits are3seconds. Whole-job elapsed includes setup, discovery, receipt computation, upload and finalization. Four logical CPUs and16.77GB RAM are recorded; sampled whole-VM peaks12.92/12.00GB and active CPU1,190.28/1,612.68seconds include containers and are not exact process peaks. Both original15-minute job limits, two shards and per-project concurrency remain unchanged.
Downloaded main browser archives independently match API SHA-256: artifact12118056529
7c4055a85197d21770fe45b8ce960de4d7fbc0da7a5b5b8a33ffd4357c788cee; artifact12118212054475b4a0873dcd039edd3a954665ed9b89ab8b7d5b73d1b0ea55dc78cad9e4db5. Collected/module manifests, case IDs/names/states and source checkouts reconcile exactly309 distinct project/file pairs,9,050 passes and the same ten existing pointer skips, without unhandled errors. Every per-file case set deeply equals the ready proof, which also deeply equals the original admitted successful measurement source. Coverage is preserved through the138/171 redistribution; neither totals nor a passing summary hides a dropped case.Main costs 9,321 summed executed job-seconds /1,238 event-to-last-job seconds; ready costs9,373 /1,176. These exclude separate draft,contract,project-sync and local work. Main's browser jobs are faster than ready while overall elapsed is longer, illustrating why the observed capacity qualification is not a whole-CI or randomized causal speedup claim. Historical cancelled identities and the independent #1229 observation retain their original attribution. No unchanged retry, extra shard, timeout increase, assertion reduction or guard/pin weakening was used.
Actual main job checkout receipts, all at
63fde3c8fe76e754aa4524f95b23f9786f4447fe:All #2366 acceptance criteria are satisfied: validated successful measurement source; browser-only refresh with all56 layout costs/provenance preserved; real complete current discovery and exactly-once partition; original limits/assertions/pins/guards/artifacts; focused92/479 controls, intended mutation/restoration and required956/6,328 preflight; fresh automatic ready/main proof with at least60seconds observed whole-job headroom; exact checkout/case/resource/queue/cost/aggregate reconciliation. No consumer package or contract change required a changeset. PR and issue retain release0.44.0.
This completes #2366. Its capacity dependency is satisfied for #1855/#1942/#2118 and their preserved draft implementations. Their own acceptance and fresh candidate qualification remain; #1942 still has its independent #2365 blocker. No dependent PR is promoted or merged by this receipt, and neither parent epic is claimed complete. This invocation's claim is released on closure; original/local evidence is retained outside the source tree.