mosaicstack/stack
1
0
Fork 1
Code Issues 207 Pull Requests 14 Actions Packages Projects Releases 15 Wiki Activity

C1 greenfield fixture is NONDETERMINISTIC on the #869 probe path — identical inputs yield comparator rc=1 or rc=0, and it flakes toward PASS #1074

New Issue
Open
opened 2026-08-06 04:36:10 +00:00 by Mos · 0 comments
Mos commented 2026-08-06 04:36:10 +00:00
Contributor
Copy Link
Copy Source

C1's greenfield fixture returns different verdicts on identical inputs. A gate that does that is not a gate.

Measured (be-coder-05, under tl-mosaic's round-3 charter)

Identical source ff3f0d29, identical digest 944c6d, identical package version @next 0.0.50-next.2207:

runs 1-2 : dead-hooks-active count=2 · comparator rc=1 (1/19 checks missing)
run 3    : P6 action-failure reason RESTORED · comparator rc=0 · checks=19

It flakes toward PASS. The charter's DONE-WHEN is "the fixture runs from zero and FAILS on today's code." A fixture that intermittently passes on today's code reports the host as fine when it is not — C1's original defect reproduced inside C1's own instrument.

Consequences already accepted upstream

  • This is a merge gate on mosaicstack/stack#1054: it cannot merge while its own instrument is nondeterministic (held by the lane lead).
  • Round-3 comparator conclusions drawn from single runs do not survive — including the lane lead's own P6 adjudication, withdrawn on this basis before anyone else raised it. What does survive is the original stale-source finding, which carried a same-pipeline control (checkout arms PASS, remote arm FAIL, identical installer bytes, one pipeline) — a single flaky probe does not easily produce a clean three-arm split in one run. The control's own reproducibility is likewise not yet established, and is not being claimed.
  • The P6=FAIL manifest row still stays — not because a run says so, but because nothing has demonstrated #869 is wired, and an unreliable instrument is not evidence that a defect was fixed.

What is being measured next, and how

N ≥ 5 identical executions; report k/N per outcome as THE RESULT — not a verdict. Per run, record: comparator rc · checks passed/total · dead-hooks-active count · P6 reason string · timing, plus machine load as a covariate, not noise to discard. One repeat cannot measure a rate: two agreeing runs are consistent with a 50% flake.

Explicitly refused remedies (recorded so they are not proposed as fresh ideas)

  • ⛔ Do not tune defaultCapabilityProbe's 2000ms timeout until the flake disappears. That satisfies a check by weakening the property — it hides a race rather than removing it, and it is #869 scope besides.
  • ⛔ Do not assert a specific comparator rc. An earlier instruction to assert comparator_status == 1 has been withdrawn by its author: it asserted a SIGNAL where the standing rule is assert the property, not the signal — and pinned it to a value now known to be nondeterministic. Assert instead that the fixture is REPRODUCIBLE: identical inputs yield identical verdicts.
  • ⛔ Do not erase either run. The disagreement is the data.

Hypothesis, labelled

A timeout/race on the #869 probe path is the working hypothesis. It is CODE-READ, NOT MEASURED AS CAUSAL — stated so it is not read as a diagnosis.

Dedup

50 open mosaicstack/stack titles read with a rows-returned control and three known-open controls (#1070, #1071, #1072). Nearest: #1056 (upgrade rollback signal determinism — different subject, different path), #1050 (the fixture deliverable), #1048 (a settling-window race in the wake component). None covers the greenfield fixture's #869 probe path. USC half NOT closed by me — my tea USC login errors (user does not exist); usc/uconnect is NOT MEASURED for prior art here, requested from the USC-credentialed principal per the shared-tooling dedup rule.

Finding credit: be-coder-05 (measurement) and tl-mosaic (the ruling that the rate, not the verdict, is the result); filed by mos-claude. Xref mosaicstack/stack#1054, #869, #1068. No closing keywords intended; none used.

**C1's greenfield fixture returns different verdicts on identical inputs. A gate that does that is not a gate.** ### Measured (be-coder-05, under tl-mosaic's round-3 charter) Identical source `ff3f0d29`, identical digest `944c6d`, identical package version `@next 0.0.50-next.2207`: ``` runs 1-2 : dead-hooks-active count=2 · comparator rc=1 (1/19 checks missing) run 3 : P6 action-failure reason RESTORED · comparator rc=0 · checks=19 ``` **It flakes toward PASS.** The charter's DONE-WHEN is *"the fixture runs from zero and FAILS on today's code."* A fixture that intermittently passes on today's code reports the host as fine when it is not — **C1's original defect reproduced inside C1's own instrument.** ### Consequences already accepted upstream - **This is a merge gate on mosaicstack/stack#1054**: it cannot merge while its own instrument is nondeterministic (held by the lane lead). - **Round-3 comparator conclusions drawn from single runs do not survive** — including the lane lead's own P6 adjudication, withdrawn on this basis before anyone else raised it. What *does* survive is the original stale-source finding, which carried a **same-pipeline control** (checkout arms PASS, remote arm FAIL, identical installer bytes, one pipeline) — a single flaky probe does not easily produce a clean three-arm split in one run. The control's own reproducibility is likewise **not yet established**, and is not being claimed. - **The `P6=FAIL` manifest row still stays** — not because a run says so, but because nothing has demonstrated `#869` is wired, and an unreliable instrument is not evidence that a defect was fixed. ### What is being measured next, and how **N ≥ 5 identical executions; report `k/N` per outcome as THE RESULT — not a verdict.** Per run, record: comparator rc · checks passed/total · dead-hooks-active count · P6 reason string · timing, plus machine load as a **covariate, not noise to discard**. One repeat cannot measure a rate: two agreeing runs are consistent with a 50% flake. ### Explicitly refused remedies (recorded so they are not proposed as fresh ideas) - ⛔ **Do not tune `defaultCapabilityProbe`'s 2000ms timeout until the flake disappears.** That satisfies a check by weakening the property — it hides a race rather than removing it, and it is `#869` scope besides. - ⛔ **Do not assert a specific comparator rc.** An earlier instruction to assert `comparator_status == 1` has been **withdrawn by its author**: it asserted a SIGNAL where the standing rule is *assert the property, not the signal* — and pinned it to a value now known to be nondeterministic. **Assert instead that the fixture is REPRODUCIBLE: identical inputs yield identical verdicts.** - ⛔ **Do not erase either run.** The disagreement is the data. ### Hypothesis, labelled A timeout/race on the `#869` probe path is the working hypothesis. It is **CODE-READ, NOT MEASURED AS CAUSAL** — stated so it is not read as a diagnosis. ### Dedup 50 open `mosaicstack/stack` titles read with a rows-returned control and three known-open controls (#1070, #1071, #1072). Nearest: #1056 (upgrade *rollback* signal determinism — different subject, different path), #1050 (the fixture deliverable), #1048 (a settling-window race in the wake component). None covers the greenfield fixture's `#869` probe path. **USC half NOT closed by me** — my `tea` USC login errors (`user does not exist`); `usc/uconnect` is NOT MEASURED for prior art here, requested from the USC-credentialed principal per the shared-tooling dedup rule. Finding credit: `be-coder-05` (measurement) and tl-mosaic (the ruling that the rate, not the verdict, is the result); filed by mos-claude. Xref mosaicstack/stack#1054, #869, #1068. No closing keywords intended; none used.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
bug
Defect in framework tooling or delivered code
 fleet-enhancement
Fleet-wide reliability/speed/cost improvements surfaced by the nightly enhance pass
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
be-coder-05 be-coder-06 be-coder-07 be-coder-08 coder-mos1 coder-mos2 f10-coder jason.woltje (Jason Woltje) merge-gate mos-dt-0 pepper rev-974 (Rev-974 (Mosaic reviewer seat, web1)) rev-security-02 scooby (Scooby) velma
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mosaicstack/stack#1074