Repository navigation
[STABILITY] Pass five consecutive literal full repository gates #48
Description
TASKS ID: STABILITY
PRD: 16 Phase 9B
Depends on all deterministic product/test rows
Acceptance:
- run literal npm run check five consecutive times on the integrated feature-freeze SHA
- record every run and exact counts
- diagnose and fix any flake causally
- no sleeps, retry inflation, worker hiding, suppression, or weakened assertions
- pass C,I
Activity
kyashp commentedon Aug 30, 2026
Dependency: run the five consecutive gates only after deterministic feature freeze, including #41–#45 where applicable.
fix(48): exit the recovery fixture after the IPC flush, not before it
sanjey99 commentedon Aug 31, 2026
Ownership — STABILITY campaign
Claiming issue #48 on branch
test/48-stability-campaign, based on currentorigin/main. I will first reproduce and causally diagnose the two remaining recordedservice.test.tsflakes, then run the literalnpm run checkgate five consecutive times on one immutable integrated feature-freeze SHA. Any head change restarts the counted sequence.Primary ownership: stability-only causal test fixes if reproduced;
apps/server/src/shepherd/service.test.tsand directly implicated fixture/support code;docs/FIXES.md,docs/BUILD_LOG.md, and theSTABILITYrow indocs/TASKS.mdafter observed evidence.Excluded/coordination boundaries: frozen UI/theme and issue #45 independent
U; issue #89 security review/posture decision; issue #65 live/model gate; unrelated product refactors, dependencies, sleeps, timeout inflation, retries, worker hiding, suppression, or weakened assertions.Acceptance plan: preserve RED evidence, fix only proven causes, pass targeted and adjacent checks, run literal
npm run checkfive consecutive times with exact counts on one SHA, obtain independent review, integrate through a linked PR, and rerun on protectedmainbefore claimingC,I.test: harden stability campaign fixtures
sanjey99 commentedon Aug 31, 2026
Draft PR #93 is open for the causal stability-fixture fixes: #93
Candidate
bd77952remains based onorigin/main@9f8579370247410a5e85df57bb4b6fe219e675b7.Observed before the formal campaign: general-project 9/9; F-03 5 consecutive focused runs at 2/2 each; PERF-01 5/5; attention cleanup 5/5; full server 844 passed / 3 skipped; production and test-source typechecks passed; dependency audit found 0 vulnerabilities. The normal pre-push gate then passed launcher 9/9, server 844/3 skipped, web 20/20, all typechecks, and both builds without reproducing outer-index staging.
The literal five-consecutive-
npm run checksequence and evidence-ledger commit start next. This issue remains open; integration and the updated-protected-main rerun are still pending.sanjey99 commentedon Aug 31, 2026
Stability campaign reset
The first formal sequence on candidate
bd77952is invalidated and reset to zero.No retry or timeout change was applied. Follow-up RED evidence showed that replacing the candidate-start polling loop with an exact promise barrier still timed out under a fresh full-suite run at 6.229s. A purpose-built
autoResolution:falsefull-Mission variant also remained over budget at 6.038s. The Git-backed Mission lifecycle is therefore the wrong five-second harness for the scheduler-overlap contract under suite pressure.The revised TDD plan moves the existing production
allSettledBoundedprimitive into the canonical scheduler module and tests the start-before-release/completion-after-release intervals directly at that exact boundary. Adjacent 30-second real-Planes tests continue to prove completed Mission, candidate selection/promotion, andpromotion_completed. No timeout, retry, worker, skip, or assertion-strength change is planned.The candidate SHA will change. All five literal
npm run checkruns will restart from run 1 after the new causal fix is reviewed and committed.test: harden stability campaign fixtures
sanjey99 commentedon Aug 31, 2026
Candidate campaign complete — integration still pending
Draft PR #93 is updated with the completed candidate evidence: #93
a2a464d6be697ab949b254944b81c355a9be540310c95da098755c513a5908615313b030bd866391ec015b6414b56df297a83489004edae441243483The invalid
bd77952sequence remained reset after its fourth run exposed the full-Mission PERF-01 timing harness at 5.176s. Four causal corrections now isolate inherited Git selectors, F-03 post-cleanup observation, scheduler overlap, and candidate-cleanup preflight. No sleep, timeout/retry increase, hidden worker, suppression, skip, or weakened assertion was introduced.Each row below was a separate literal
npm run checkon the same candidate and base:Hosted candidate runs
33351880563and33351880571passed. The documentation-only final head also passed the complete local pre-push gate, then exact-head hosted runs33353238685(2m54s) and33353238734(3m43s) passed.Candidate gate
Cis passed. This issue remains open and the PR remains draft: maintainer/integrator review, protected-main merge, and a fresh gate on the resulting protected-main SHA are still required for gateIand closure.sanjey99 commentedon Aug 31, 2026
Post-merge closeout ownership
PR #93 is merged at protected-main SHA
b05a4aad10d0c4c0e9f7d54fed2cdd5c779c16b0. I am taking the bounded integration-evidence closeout on branchdocs/48-stability-integration-evidencefrom that exact SHA.Owned files: the
STABILITYrow indocs/TASKS.md, issue #48 entriesTST-27–TST-30indocs/FIXES.md, and a bounded exact-main entry indocs/BUILD_LOG.md.Excluded: product/test/UI code, dependencies, other ledger rows, issue #45 usability, issue #89 security, issue #65 live/model work, and all unrelated documentation.
The exact merge has already passed one fresh literal local
npm run checkand hosted protected-main run33353622211. This closeout branch will record only those observed results; it will not alter the verified behavior.docs: record issue 48 integration evidence
sanjey99 commentedon Aug 31, 2026
Protected-main gate
IpassedPR #93 merged as
b05a4aad10d0c4c0e9f7d54fed2cdd5c779c16b0; its tree is byte-identical to final PR headec015b6.npm run check: PASS / exit 033353622211, PASS on exactb05a4aaCandidate gate
Cand integration gateIare both passed. The bounded ledger closeout is ready in PR #95: #95PR #95 changes only
docs/BUILD_LOG.md,docs/FIXES.md, and theSTABILITYrow indocs/TASKS.md; its own hosted runs33354236483and33354236478passed. It containsCloses #48, so merging it will integrate the evidence and close this issue without a direct protected-main write.