The second follow-up review found no Blocker and no Major, and four
Minors that are corrected. Test-MatrixCleanup.ps1 and the README say
that every unresolved S-1-5-21-* member of Performance Log Users counts
as an entry of the probe; a Verify with the new script found none on
the two machines of the first lab. The record says that the replay
shows that the module doesn't decide the outcome and that six runs can't
rule out a small effect, which the result bullet and the summary of the
evidence had left out; it lists the first-lab counts among its tables,
names the controller blob of rc7d, says that the first-lab check ran
before the profile and log-group fields existed, counts nine restarts
inside the series, and mentions the first attempt of the cleanup test
that died. The controller comment says "for the baseline and for the
final candidate alike" instead of "whichever version"; no code changed.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: AI Assistant <ai@example.com>
The follow-up review found no Blocker and no Major, and five Minors, all
verified against the raw logs and all corrected: rc7e's first cell was
4.6 minutes after the removal of rc7d, which created and removed
accounts without a test (the timeline now includes rc7d); one lifetime
that fits both tests is 9.95 to 10.25 minutes, not 9.35 to 10.20; the
client restarted at 05:34, which matters for the counterfactual of ab7
only; and five statements that the record's own data contradicted
(the age of the entries in failing and passing cells, the blob of the live
tests of the replay, the exporter's scope, the folders that the check
counts, and the blob of the controller of ab7 to ab10).
The record says that six replay runs cannot rule out a small effect of the
module, that the window of the client test is 0.3 minute wide, and that
the cleanup check ran with real residue of every kind.
The final candidate also passed the live controller in the first lab,
where case 9 runs: 245 passed, 0 failed, 1 skipped per edition, with the
fixture removed and verified clean.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: AI Assistant <ai@example.com>
The effective-access failures of the Admin role in the Windows Server
2022 cell don't depend on the module. A replay of seven cells with the
baseline and the final candidate alternating failed the baseline in two
of three cells and the final candidate in one of three, not counting the
warm-up. In a failing cell the remote authorization managers answer as
if the account had no groups while the name resolution, the Kerberos
logon, and the local manager are right in the same second.
One model with one lifetime (9.35 to 10.20 minutes) fits all 43 Admin
role runs of 27 cells, and none of 5,000 random assignments of the
outcomes does. Four more cells with unique account names pass, two of
them where the model predicts a failure for a reused name.
The record, the README, and a timeline CSV now say what the evidence
supports and what it doesn't establish, that the cleanup check ran with
real residue, and that fdd7a8b reverts cleanly while 962887a conflicts
with it.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: AI Assistant <ai@example.com>
The record, its tables (the suite per candidate, the failing tests, and the
controller cells), and the README of the lab. The final local candidate passes the
module's suite on Windows Server 2019, 2022, and 2025, Windows 11 22H2 and 26H1,
and the host in both editions, elevated and as a basic user, and the live
controller in three cells. The record states what a local build does not prove.
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: AI Assistant <ai@example.com>
Raimund Andree