From 9b9e40df690c70a0271051a9957fc5168459f7a8 Mon Sep 17 00:00:00 2001 From: Luke Curley <432854+kixelated@users.noreply.github.com> Date: Fri, 16 Oct 2026 10:53:38 -0700 Subject: [PATCH 1/5] chore: claim deterministic TS recovery quest From 98562b69d04e377613166653689b8430a43f5c02 Mon Sep 17 00:00:00 2001 From: Luke Curley <432854+kixelated@users.noreply.github.com> Date: Fri, 16 Oct 2026 10:58:33 -0700 Subject: [PATCH 2/5] docs(quest): gate TS recovery on whole-group acquisition Co-Authored-By: GPT-6 --- quest/m1/ts-reliability/recovery.md | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/quest/m1/ts-reliability/recovery.md b/quest/m1/ts-reliability/recovery.md index edce1671009..9ad0349fa1f 100644 --- a/quest/m1/ts-reliability/recovery.md +++ b/quest/m1/ts-reliability/recovery.md @@ -9,6 +9,15 @@ missing dependency chains are not presented as independently decodable. ## Plan +The cold-acquisition baseline is blocked by [#5713](https://github.com/moq-dev/moq/pull/5713). +Reproduced on `main` at `e6006793f` on 2026-10-16 with that PR's +`a_start_group_cut_short_starts_at_the_next` paused-clock regression, omitting +only the counters that the PR adds: after the first group is reset, export +still writes its five pictures at PTS 0 through 14400 before the next whole +group at PTS 54000. The oracle requires starting with that whole group. +Keep the production fix and its regression with the source quest below; +resume the broader recovery evidence after it merges. + Test content loss separately from IP packet loss. QUIC may retransmit IP loss; a netem percentage does not establish that any MoQ object was omitted. Keep the real network impairment arm for transport and pacing evidence, and @@ -40,6 +49,10 @@ Short deterministic arms run on relevant PRs. Long drift and impairment arms run nightly with failure artifacts. Keep byte-position PCR accuracy and measured release timing as separate checks. +## Required + +- [TS decoded content](/quest/m1/test-flakes-2/ts-video-content.md) - a reset acquisition group is discarded before TS output begins + ## Related - [Automated validation](/test/ts/README.md) - shared graders, profiles and known-bad controls From 187c85a639d1b611e2aee0771c84f3d98133ba06 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 17:30:31 -0700 Subject: [PATCH 3/5] quest(m1): build TS recovery on #5713's whole-group start #5713 merged with the cold-join regressions this draft gated on, so drop the prerequisite (its quest is gone) and record what the recovery tests reuse instead. Co-Authored-By: Claude Opus 5.5 --- quest/m1/ts-reliability/recovery.md | 21 +++++++++------------ 1 file changed, 9 insertions(+), 12 deletions(-) diff --git a/quest/m1/ts-reliability/recovery.md b/quest/m1/ts-reliability/recovery.md index 3c35d7c41c8..dff9c02d58e 100644 --- a/quest/m1/ts-reliability/recovery.md +++ b/quest/m1/ts-reliability/recovery.md @@ -9,14 +9,14 @@ missing dependency chains are not presented as independently decodable. ## Plan -The cold-acquisition baseline is blocked by [#5713](https://github.com/moq-dev/moq/pull/5713). -Reproduced on `main` at `e6006793f` on 2026-10-16 with that PR's -`a_start_group_cut_short_starts_at_the_next` paused-clock regression, omitting -only the counters that the PR adds: after the first group is reset, export -still writes its five pictures at PTS 0 through 14400 before the next whole -group at PTS 54000. The oracle requires starting with that whole group. -Keep the production fix and its regression with the source quest below; -resume the broader recovery evidence after it merges. +[#5713](https://github.com/moq-dev/moq/pull/5713) settled the cold-join +start: an export begins on a whole group, and `ts::stats::Export` counts +`trimmed`, `gaps` and `dropped`. Its paused-clock regressions +(`a_start_group_cut_short_starts_at_the_next`, `a_join_drops_its_group_whole`) +already cover a start group cut before output, so this quest does not repeat +them. Extend the evidence to loss after output began and to decoded output, +and check each fault schedule against those counters: a hole the oracle +accepts must be one the export reported. Test content loss separately from IP packet loss. QUIC may retransmit IP loss; a netem percentage does not establish that any MoQ object was omitted. @@ -49,11 +49,8 @@ Short deterministic arms run on relevant PRs. Long drift and impairment arms run nightly with failure artifacts. Keep byte-position PCR accuracy and measured release timing as separate checks. -## Required - -- [TS decoded content](/quest/m1/test-flakes-2/ts-video-content.md) - a reset acquisition group is discarded before TS output begins - ## Related - [Automated validation](/test/ts/README.md) - shared graders, profiles and known-bad controls +- [Audio lead](/quest/m1/ts-reliability/audio-lead.md) - audio stays behind the keyframe when the start group is dropped - [Gradual recovery](/quest/m2/intra-refresh/README.md) - recovery counts and per-group warmup From b4a43a23527e71dbd579f37ff9b89b046b244f79 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 18:39:11 -0700 Subject: [PATCH 4/5] quest(m1): drop the audio-lead link, fixed by #5810 Co-Authored-By: Claude Opus 5.5 --- quest/m1/ts-reliability/recovery.md | 1 - 1 file changed, 1 deletion(-) diff --git a/quest/m1/ts-reliability/recovery.md b/quest/m1/ts-reliability/recovery.md index dff9c02d58e..ed310bf27cf 100644 --- a/quest/m1/ts-reliability/recovery.md +++ b/quest/m1/ts-reliability/recovery.md @@ -52,5 +52,4 @@ and measured release timing as separate checks. ## Related - [Automated validation](/test/ts/README.md) - shared graders, profiles and known-bad controls -- [Audio lead](/quest/m1/ts-reliability/audio-lead.md) - audio stays behind the keyframe when the start group is dropped - [Gradual recovery](/quest/m2/intra-refresh/README.md) - recovery counts and per-group warmup From 859e45346b4a6dc5d5ca7165e8d80f1987fdbd68 Mon Sep 17 00:00:00 2001 From: Luke Curley Date: Fri, 16 Oct 2026 18:59:31 -0700 Subject: [PATCH 5/5] quest(m1): grade a missing track by its stream liveness Co-Authored-By: Claude Opus 5.5 --- quest/m1/ts-reliability/recovery.md | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/quest/m1/ts-reliability/recovery.md b/quest/m1/ts-reliability/recovery.md index ed310bf27cf..6b906453149 100644 --- a/quest/m1/ts-reliability/recovery.md +++ b/quest/m1/ts-reliability/recovery.md @@ -15,8 +15,9 @@ start: an export begins on a whole group, and `ts::stats::Export` counts (`a_start_group_cut_short_starts_at_the_next`, `a_join_drops_its_group_whole`) already cover a start group cut before output, so this quest does not repeat them. Extend the evidence to loss after output began and to decoded output, -and check each fault schedule against those counters: a hole the oracle -accepts must be one the export reported. +and check each fault schedule against what the export reported: a hole the +oracle accepts must show in those counters, or for a track that delivers +nothing, in its stream row's `units` and `quiet`. Test content loss separately from IP packet loss. QUIC may retransmit IP loss; a netem percentage does not establish that any MoQ object was omitted.