[kixelated]

Problem

Investigate the reported publisher-disconnect truncation in /quest/m1/test-flakes-2/subscription-cut.md: a subscriber sometimes ended cleanly after 10 of 20 frames.

Approach

This draft records the claim and blocked investigation; it contains no source changes and preserves the quest. Eight full moq-tokio suite runs on the current questline base passed all 426 tests each (3,408 passes total), including the disconnect regression every time. Command: CARGO_BUILD_JOBS=4 NEXTEST_TEST_THREADS=4 nix develop --command just rs test -p moq-tokio --retries 0.

Merged #4533 already hardened active-stream tail settling since the original report. Its description also identifies aborted groups being skipped by arrival readers, which /quest/m1/subscribe-drop.md tracks. The ongoing session-death work fixes peer-error propagation and local-close behavior while preserving settlement semantics. No remaining root cause was reproduced, so no speculative fix was made.

Impact

• Public API: none.
• Wire: none.

Alternatives

Changing terminal-state policy without a reproducer would overlap the existing reader-end and SUBSCRIBE_DROP quests and could change intentional gap semantics.

Follow-ups

Reproduce on the current source under the original broader workspace load, or recheck after the overlapping fixes land. Recommendation: leave this draft until there is a failing regression or evidence that the quest is already resolved. Saved decisions for the user's return.

(Written by GPT-6)

[kixelated]

Investigation outcome: eight complete moq-tokio suites passed (3,408 tests, disconnect regression passed 8/8) on the current questline base. No code changes made; quest remains open. The original broader-workspace load has not been reproduced. Existing #4533 addresses tail settling, while aborted-group visibility and terminal semantics overlap SUBSCRIBE_DROP/reader-end-parity work. Recommendation: keep this draft, recheck after those changes, and require a current failing regression before changing behavior. No new quest proposed. Decisions deferred until the user returns.

(Written by GPT-6)