[kixelated]

Finishes quest/m1/lite-timestamps-through.md (deleted here, with its links).

Problem

moq-dev/moq#6150 made receivers drop every timestamp read over moq-lite-05/06, because an untimed origin stamps send times there. As a result, a lite-06 reader of a timed feed (such as a stats feed) saw untimed frames. moq-dev/moq#6300 also served moq-transport tracks to moq-lite untimed unless the ietf extension was negotiated. Maintainer decision (2026-10-25): pass the timestamps through instead.

Approach

• Rust: track::Info gets back a crate-private unchecked flag. The lite subscriber sets it on a lite-05/06 copy. TrackState::neighbors skips the moq-dev/moq#5803 rewind check for unchecked or moq-transport-provenance tracks. Frame, fetch, and datagram ingest keep the timestamps they read.
• Publishing: Info::lite_timescale and group::Consumer::lite_timescale are gone. Every lite hop serves the track's own timescale and timestamps, including to TRACKS_ADD, lite-07 without the extension, and lite-05/06. An untimed track on lite-05/06 still carries its send time in milliseconds, as published.
• JS: @moq/net mirrors this. The @internal track.Info.ietf becomes unchecked. lite-05/06 TRACK_INFO maps to a timed copy with unchecked. frameDecoder/readFrames lose their timed flag, and liteTimescale is gone.
• Specs and docs: draft-lcurley-moq-lite (receiver text and changelog), draft-lcurley-moq-ietf (timestamps without the extension), doc/concept/moq-lite.md, and doc/setup/upgrade.md.
• Tests: they reverse feat(lite): drop lite-05/06 timestamps so receivers check received groups #6150's expectations in moq-net, moq-tokio, moq-relay, moq-cli, the C/C++ probes, e2ee, and test/wasm. received_rewind.rs now covers:
  • a timed track crossing a lite-06 (and moq-transport) hop between two origins with its values, served on over lite-06, lite-07, and moq-transport;
  • send times from an untimed origin passing a lite-06 relay unchecked, with lite-07 refusing the out-of-order group;
  • out-of-order fetches on lite-05/06 being accepted with their send times.
• Interop finish clients: they no longer assert that the .health frame is untimed. A lite-06 browser publisher's untimed health track now reaches every reader with send times.

Choices made without asking

• track::Info equality (Rust PartialEq, JS resume compatible) ignores unchecked. It records how a copy arrived, not what it holds, so same-epoch copies over lite-06 and lite-07 still splice. Before this change they mismatched on timescale.
• The JS Info.ietf field (@internal, unreleased) is renamed to unchecked. JS used it for nothing else.
• has_media_time stays as the version predicate. It now means "set unchecked", and resolves_anchors keeps requiring it.

Impact

• Public API: none in Rust (unchecked is crate-private). In JS, the @internal track.Info.ietf becomes @internal unchecked.
• Wire: no format change. Receiver behavior changes:
  • A track received over moq-lite-05/06 arrives timed with the TRACK_INFO timescale. That is the publisher's media time, or send times in milliseconds for an untimed origin. Its groups are not rewind-checked.
  • A relay serves those, and moq-transport timestamps, on over every version. A lite-07 receiver still checks them, so a send-time origin's out-of-order group (or a moq-transport rewind without the ietf extension) may be refused downstream. This cost was accepted in the quest.
• Migration: feat(lite): drop lite-05/06 timestamps so receivers check received groups #6150 and fix(net): preserve IETF timestamps across negotiated lite hops #6300 are unreleased, so against the last release this restores the lite-05/06 pass-through. The existing Unreleased MIGRATE.md notes from feat(lite): drop lite-05/06 timestamps so receivers check received groups #6150 (rs/moq-net and js/net: "A track received over moq-lite-05/06 arrives untimed…", "a relay serves it to moq-lite untimed") are now wrong. As the quest plans, the cut's migration audit reconciles them with this section: received lite-05/06 tracks keep their timestamps unchecked, and moq-transport timestamps cross moq-lite timed.

Alternatives

• A wire flag telling lite-07 receivers a track is unchecked. The quest rejected it: rewinds are unsupported.
• Keeping Info equality derived. Simpler, but it would refuse a same-epoch lite-06 copy beside a lite-07 one.

Checks

just ci passes. just test interop --all passes: the full matrix ran, and the finish matrix was rerun after the health-assertion fix.

Follow-ups

None found.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)