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
Issue #811 closed with four of its five acceptance criteria checked. AC4 is still unchecked on main:
docs/features/active/2026-09-07-utilitiescs-test-determinism-780-803-594-811/spec.md:295
- [ ] AC4: A full nine-assembly /InIsolation run with TestCategory!=LiveOutlook reports zero
failures on ten consecutive runs, recorded as evidence.
AC1, AC2, AC3 and AC5 are all [x]. AC4 is the gate that proves the determinism work actually eliminated the races rather than merely suppressing them, so leaving it open leaves the issue's central claim unverified.
AC4 was deferred for a stated reason, not overlooked. The remediation cycle 1 exit reaudit recorded that AC4 should be re-evaluated once the ILGlobals race was fixed under its own issue, because that race was a known source of cross-class interference in a full-suite run. That precondition is now discharged: issue #824 was delivered and closed, and its feature audit is on main at docs/features/active/2026-09-08-ilglobals-loadopcodes-unsynchronised-static-race-824/feature-audit.2026-09-09T22-40.md.
Nothing has re-evaluated AC4 since. This entry exists so that re-evaluation is tracked somewhere durable rather than living only in a closed feature folder.
Implementation Intent
Re-run the AC4 gate against current main now that #824 has shipped, then either check AC4 off with the evidence attached, or record a specific, reproducible failure that justifies keeping it open and names what still needs fixing.
The AC4 gate as spec.md states it: ten consecutive full-suite /InIsolation runs over the assembly set the mstest-coverage workflow discovers (every *.Test.dll under the debug output tree, excluding intermediate and reference directories), filtered with TestCategory!=LiveOutlook, all reporting zero failures, recorded as committed evidence.
Acceptance Criteria
Ten consecutive full-suite runs are executed against current main under the filter and isolation mode spec.md names, and each run's result is captured as committed evidence.
AC4 in docs/features/active/2026-09-07-utilitiescs-test-determinism-780-803-594-811/spec.md is either checked off with a citation to that evidence, or left unchecked with a recorded failure naming the specific test and the specific interference.
If any run fails, the failing test and its suspected cause are captured as a separate potential entry rather than fixed under this one.
Dependencies / Risks
This is an evidence-gathering task, not a code change. If it passes, the only edit is a checkbox and an evidence file. Resist widening it into a fix; a failure belongs in its own issue.
Two known local-environment hazards will distort the result if ignored. Four UtilitiesCS.Test shell-icon test classes stall vstest on at least one development machine through SHGetFileInfo, reproducing on main and unrelated to Bug: utilitiescs-test-determinism-780-803-594 #811; and a local run needs both the \.claude\ worktree-path exclusion and CI's /InIsolation to avoid loading stale assemblies out of agent worktrees. A run that omits either produces failures that look like determinism regressions and are not. CI is the more trustworthy venue for this gate.
Ten consecutive runs is a long wall-clock commitment. Consider whether CI can carry the repetition rather than a developer machine.
Problem / Why
Issue #811 closed with four of its five acceptance criteria checked. AC4 is still unchecked on
main:AC1, AC2, AC3 and AC5 are all
[x]. AC4 is the gate that proves the determinism work actually eliminated the races rather than merely suppressing them, so leaving it open leaves the issue's central claim unverified.AC4 was deferred for a stated reason, not overlooked. The remediation cycle 1 exit reaudit recorded that AC4 should be re-evaluated once the
ILGlobalsrace was fixed under its own issue, because that race was a known source of cross-class interference in a full-suite run. That precondition is now discharged: issue #824 was delivered and closed, and its feature audit is onmainatdocs/features/active/2026-09-08-ilglobals-loadopcodes-unsynchronised-static-race-824/feature-audit.2026-09-09T22-40.md.Nothing has re-evaluated AC4 since. This entry exists so that re-evaluation is tracked somewhere durable rather than living only in a closed feature folder.
Implementation Intent
Re-run the AC4 gate against current
mainnow that #824 has shipped, then either check AC4 off with the evidence attached, or record a specific, reproducible failure that justifies keeping it open and names what still needs fixing.The AC4 gate as
spec.mdstates it: ten consecutive full-suite/InIsolationruns over the assembly set themstest-coverageworkflow discovers (every*.Test.dllunder the debug output tree, excluding intermediate and reference directories), filtered withTestCategory!=LiveOutlook, all reporting zero failures, recorded as committed evidence.Acceptance Criteria
mainunder the filter and isolation modespec.mdnames, and each run's result is captured as committed evidence.docs/features/active/2026-09-07-utilitiescs-test-determinism-780-803-594-811/spec.mdis either checked off with a citation to that evidence, or left unchecked with a recorded failure naming the specific test and the specific interference.Dependencies / Risks
UtilitiesCS.Testshell-icon test classes stallvsteston at least one development machine throughSHGetFileInfo, reproducing onmainand unrelated to Bug: utilitiescs-test-determinism-780-803-594 #811; and a local run needs both the\.claude\worktree-path exclusion and CI's/InIsolationto avoid loading stale assemblies out of agent worktrees. A run that omits either produces failures that look like determinism regressions and are not. CI is the more trustworthy venue for this gate.TryAddValuesAsynctest is tracked separately under issue Bug: tryaddvaluesasync-wall-clock-timeout-flaky #780.Verification Steps
/InIsolationwithTestCategory!=LiveOutlook, repeated ten times.Evidence Checklist
Source
From: docs/features/potential/2026-09-11-issue-811-ac4-ten-run-determinism-evidence-outstanding.md