Repository navigation
quest: plan what the pre-cut audit decided - #8892
Merged
Merged
Conversation
An audit of everything on main since release, and of the lite-07 draft, ahead of the 0.22 cut. The maintainer decided each finding; this records them as m2 quests, moves and deletes the quests they supersede, and rewords two AGENTS.md rules. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
kixelated
marked this pull request as ready for review
November 16, 2026 09:41
kixelated
enabled auto-merge
November 16, 2026 09:41
Contributor
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes: the PR's non-quest edits in full, plus a reference check across the quest tree. I did not read each quest document's content.
AGENTS.mdrules: two reworded rules, one for the leading.path segment and one for when a feature reaches the bindings.- Removed ignored test:
a_relay_with_holes_answers_a_wide_past_range_plus_live_over_moq_transportand itssh/gh/test-inventory.allowline are both gone, so the inventory stays consistent. - Interop skip text: the
test/interop/interop.shskip detail andtest/interop/README.mdno longer point at the deletedietf-finishquest. - Quest tree: about 130 quest documents added, edited, moved or deleted. No dangling links to the deleted quests remain, only prose that records the deletions.
claude-sonnet-5-5 | 𝕏
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.

Problem
The 0.22 cut publishes every Rust crate, JS package, and binding as a semver break, and freezes the lite-07 wire. After it, a removal or rename costs another breaking release and a wire change costs lite-08. The 2026-11-14 feature review judged about 50 work lines "ship as is". The maintainer asked for an exhaustive audit of the same window before the announcement: is each addition worth it, made for a long-term reason, named well, and as simple as it can be, and can lite-07 itself lose anything.
Approach
The audit covered
d0b086e71b(thereleasemerge base) to6ac78b68ee, about 3,800 commits: a digest of moq.pro's and Kixel's roadmaps, fifteen read-only reviewers by area (about 480 work lines), my own check of the claims that became questions, and a 15-round interview in which the maintainer decided each finding. This PR writes those decisions into the quest tree. It changes no shipped code.Every quest's Plan records its decision as
✅ chosen (rejected: ...), dated 2026-11-16, with the evidence.quest/m2/README.mdhas the index.Existing quests:
m3/time-addressing-consumers(nowm2/time-addressing/consumers) andm4/ietf-time-addressing. Moved to m5:m3/ts-multi-program-restart. Renamed:m3/release/demo-sockets-deploytodemos-deploy.m2/subscriber-min-age(reversed),m3/ietf-finishandm3/ietf-hole-stall(moq-transport interop polish between our own clients),m3/lite-finished-across-updates(its premise was the reopen that yield-defer deletes), andm3/release/libmoq-retirewithlibmoq-final-release(no stub:libmoqships the real API).m3/webrtcand vendor-sync quests (replanned for upstream str0m), them5/e2eeandm5/net-wrapepics (crate renamed, format quest required),m3/ffi-exporter, the C release quests, and the intra-refresh epic.Outside
quest/:AGENTS.md: two rules reworded, both agreed in the interview. A feature reaches the bindings when a binding user needs it (it read "keep all languages and wrappers at the same functionality", which mirrored unused features into every binding). A path segment's leading.is the one marker allowed inside a value.ietf-hole-stallandietf-finishleft an ignored test and an interop skip naming them. The test (a_relay_with_holes_answers_a_wide_past_range_plus_live_over_moq_transport) and itstest-inventory.allowline are removed, and the skip intest/interop/interop.shand its README says the gap is not planned work.Decisions
All 2026-11-16, maintainer. The headline ones, with what they reverse:
Frame Start, the Padding stream and datagram types, and the count on SUBSCRIBE_DROP. It keeps datagram tracks with their lite-05/06 bridge, all seven lite versions, the hidden.prefix, the route plan, and Subscriber Min Age, which reverses 2026-11-14 (#8839 closed) and gains a reader. Theietfextension stays but costs nothing on a moq-lite track. The\escape is finished, not removed.moq-e2eecrate again, unpublished untilmoq-e2ee-01is finished), videowarmupin the catalog, the generated C API (libmoq0.22 is the hand-written API's final line, reversing 2026-10-25), and Dart on the shared version.One decision was made on a wrong fact, mine. I asked what to do with #8857 as an open PR ("delete the path, park #8857" was picked). It had merged ten hours earlier. video-warmup records the pick as it applies now, deleting the one-decoder path #8857 landed, and tells its PR to put this at the top of its body for the maintainer to confirm or reverse. catalog-warmup restores it from history with the producer.
Choices in the quests that the maintainer did not make are written as recommendations, not decisions. The larger ones: old config spellings are refused rather than aliased;
reserve_unlisted_trackbecomes public for Kixel's socket port; reconnect needs three states (unset, off, a value) where "absent" already means default; the hole wait's live estimate lives on the playhead, notDelay::min;Read::Endreports every group's declared end; the errored group end isfail(err).Impact
Public API: none. Wire: none. This PR plans; each quest states its own break.
The cut now waits on 53 more m2 quests (51 new, 2 moved in): 6 L, 32 M, 14 S, and 1 XS. That is the cost of the decisions above.
Follow-ups
For other repositories' quest trees, to plan there:
moq_net::socketand its wrappers from the socket removal PR's parent commit (the port inherits the closed-endpoint race inquest/m0/test-flakes-3/py-socket-abort.md); movePublicKeyinto kixel-trust (itsmoq_crypto::sign::PublicKeyimport breaks at the cut either way); check each unused-API sweep for a caller before it lands.media.clocktosync.clock; discard or rewrite any staging archive written by amainbuild (the format ships as version 2); the auth server handles strict grants, thelinkenum, and the attached-token change; check each unused-API sweep for a caller.Seen but not changed: groups-never-overlap (planned 2026-11-15 in m3) tightens the hang draft's group rule after hang-04 would publish, so it may belong in m2.
Validation
quest checkandquest check --base origin/main: 542 documents ok.just checkon the branch: passed (rs check and test with 12,825 tests, js, markdown, gh, tooling, shell, quest), which covers the removed test and the interop script.Models
Size: L
Implementer: Claude Code, claude-fable-5-2, unknown effort
🤖 Generated with Claude Code
(Written by Claude Fable 5.2)
Review
Rewritten by
quest reviewafter each round.Head: 39e9acd
Base: origin/main at 0019174
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, xhigh effort
Session:
01a1e3b8-194d-7501-b7d6-05e4366da9aa, round 1Score: 10/10,
The change records planning decisions without changing runtime behavior, and no blocking defects were found.Rounds: 1; first round 0; findings 0: fixed 0, refuted 0, open 0
Findings: none