Repository navigation
quest: browser audio takes its holes from the ordered reader - #8989
Conversation
Maintainer decision during /quest-run: the transport decides an audio hole through an ordered reader whose budget is the ring's size, replacing the measure-the-playhead plan. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes: the full diff, two quest docs, with no code touched.
- Quest rewrite:
browser-audio-hole-skip.mdnow plans browser audio reading through@moq/net's ordered reader, with a ring-sized budget andGapplayed as the hole. It also adds a Waiting entry on #8910. - Related link:
delay-hole-wait.mdpoints at the retitled quest.
The linked quest/m2/unused-api/net.md exists.
claude-sonnet-5-5 | 𝕏
kixelated
left a comment
There was a problem hiding this comment.
Automated review by review (OpenAI) · c52e9bc8bfada1c90cbb81fdcbeed5840cff877a
Please explicitly preserve pre-skip jitter observations in the plan. The container's frame ingress currently feeds the jitter estimator before gap handling or age filtering (js/hang/src/container/consumer.ts, at #jitter.observe). Once audio reads through a narrow ordered reader, a group may still reach the broad subscription but be skipped before reaching the container. Keeping the two budgets separate on paper is not enough if the estimator sees only ordered output: it can stop learning exactly when arrivals get late. Preserve a pre-skip observation path with original arrival times. A deterministic replay case should prove that a group outside the ring budget still raises the estimate without being played.
The transport owning the hole decision looks reasonable. Static review of the plan; no tests run.
|
No actionable comments were generated in the recent review. WalkthroughThe browser audio proposal now specifies using Merge Risk:
|

Summary
Rewrites
quest/m2/audio-playout/browser-audio-hole-skip.mdafter a maintainer decision made during /quest-run. Browser audio will read its track through@moq/net's ordered reader. The reader's budget is the amount of audio the ring holds, and eachGapthe reader reports is played as a hole. The audio container consumer then drops its own hole wait and stops takingplayhead. Video keeps the playhead.This replaces the quest's previous plan, which was to measure the playhead rule and then hand-write a wait in the container consumer.
The quest now includes mirroring
Item/Gapin@moq/net. #8910's own follow-up list deferred that until a consumer needs it, and browser audio is that consumer. The quest waits on #8910 merging.quest/m2/delay-hole-wait.md's Related line was updated to match.Decisions
Validation
quest checkpasses. Documentation-only change.🤖 Generated with Claude Code
(Written by Opus 5.5)
Review
Rewritten by
quest reviewafter each round.Head: c52e9bc
Base: origin/main at 60dfbbf
Risk: medium,
the base has no RISK.md to name what is always high or customer-facing, so every change is mediumClassified: medium by default
Reviewer: Codex, gpt-6.1-sol, high effort
Session:
01a1e85c-c142-74b3-a37a-4ba9250a2440, round 1Score: 10/10,
The planning-only change consistently records the settled approach without modifying runtime behavior, and no actionable defects were found.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none