Skip to content

random seed: a daemon pause that cannot engage: D_ARR not-observed fails on PR 99 at seed -1120189360, passes on main #184

Open

Description

@KyleAMathews

CI run 44588135077 (PR #99 head 44c0dbd, job "Gap report end-to-end oracle (random seed, 3/4)") failed the random unit random seed: a daemon pause that cannot engage:

ORACLE_SEED=-1120189360 ORACLE_PATH=0 ORACLE_PROPERTY='oracle-gaps.report' ORACLE_UNIT='random seed: a daemon pause that cannot engage' vitest run -t '^oracle-gaps\.report(?: | > )random seed: a daemon pause that cannot engage$'
Difference: src/sut.ts:D_ARR:31:arithmetic-swap#0: not-observed, owed indeterminate

Replay results (GAP_E2E_SHRINKS=0):

  • PR 99 at 97ddc05: FAILS twice, same Difference (17 s, 18 s).
  • origin/main at 6d05f05: PASSES twice (19 s, 18 s).

Caveat: the two trees' generators differ, so the same seed may draw different histories. Main passing produces no History: line, so the drawn main history could not be captured and compared. The CI log truncates the failing history at 2,000 characters.

Relation to #105: #105 reports the determinism unit of the same constructed history (a daemon pause that cannot engage) giving different traces across runs. This is the random-seed unit, with law classification and a not-observed mutant, so it may be a separate defect.

Needed to attribute: capture the full history on both trees (scratch uncapping of reportText in src/oracle-property.ts, reverted), compare the drawn histories field by field, and if they match, reproduce the Difference under the frozen history on main.

Activity

  1. KyleAMathews commented on Nov 19, 2026

    @KyleAMathews
    ContributorAuthor

    Replay by history (M, "Replay by history") for this failure: NOT achieved, so this issue is NOT attributed to main or to #99 yet.

    Consequence: main cannot be shown to pass or fail this history, so the failure is neither attributed to main (no main fails the same way) nor to #99 (no main passes and #99 fails on the same history). Keep it open. Blocking for #99 until one of: a replay path that takes #99's history on main (which needs #99's grammar fields there), or the history reproduced on #99 by an oracle-first witness.

    Related: #105 (same history, determinism unit, different law), #185 (2,000-char cap).

  2. KyleAMathews commented on Nov 19, 2026

    @KyleAMathews
    ContributorAuthor

    Attribution (coordinator ruling: a history with PR-only fields is the PR's own unless a neutralized form also fails on the base).

    1. PR 99 at 4c29fa8, the history with heavy: 0 and hangMs: null (its neutral values): still FAILS, twice, Difference: src/sut.ts:D_ARR:31:arithmetic-swap#0: not-observed, owed indeterminate. So the failure does not depend on heavy or hangMs.
    2. origin/main at 6d05f05, the same history with the PR-only keys dropped and their neutral values (heavy 0, hangMs null, loops 0, timerWait null, pulse false, pendingIo null, slowMs unused since no test is slow). stopRuns and laterSteps use main's constant objects (fast-check takes back constants by reference). Replayed on main as a scratch constructed entry, uncommitted, twice: FAILS both times with the same Difference: src/sut.ts:D_ARR:31:arithmetic-swap#0: not-observed, owed indeterminate.

    Ruling: the failure reproduces on main with the history's PR-only fields neutral. It is main's failure, not the PR's. Recorded in the #99 body as main's. The earlier "main passes at this seed" result was a different draw (the trees generate different histories for the same seed), so it is not evidence.

    Also: the history had to be re-expressed, so this is a neutralized replay, not the byte-identical history. The reportText cap (#185) is why the full history was only recovered by a scratch edit.

    Related: #105 (the determinism unit of this history) and #185 (history cap).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions