[kixelated]

Announcements now report availability. When this side lets go of a route, readers get a distinct offline event instead of the origin's END. This covers Rust, js/net, and every binding. Implements quest/m2/cluster-routing/routes/availability.md, building on the Listing API.

Draft, waiting on one maintainer decision (below). Unit and integration tests pass and just check passes, but every finish X -> Y lane of just test interop --all fails: a cleanly finished broadcast reaches readers through a relay as Offline, not Absent.

Waiting

Maintainer: choose how a token-scoped lite-07 session certifies withdrawals, so a clean finish ENDs across a relay. The quest file keeps this item in its own ## Waiting section.

Scoped lite-07 sessions never certify, so no END crosses a relay

The router reports END only for a claim a floor certifies. Anything else is let go and goes offline. That rule predates this PR. Before this PR, every unserve was reported as END anyway, so the gap never showed.

A token-scoped publisher answers the relay's request for "" with a floor that certifies nothing, because its grant (finish-…/**) does not cover "" (Answer::certified). The relay answering a scoped reader hits the same thing: the reader's limit doesn't cover "", and the relay only knows the origin beneath finish-…/. Every token-scoped publisher and reader has this shape, so on moq.pro every clean finish would read as offline.

What I tried and reverted: certify "the requested prefix as the grant shows it", with the router passing per-region floors to the answer. That fixed interop, but it's unsound. A receiver doesn't necessarily mirror the presenter's grant: a relay with no limit has scope None. A grant shrink then removes a claim without bumping the revision, so the receiver keeps the stale claim. auth::lite_07::tokens_union_and_withdrawing_one_shrinks_it caught it.

Options:

1. (Recommended) The subscriber narrows its ANNOUNCE_REQUEST to its scope's single region. That region is where its limit and its subscribe grant meet. It reopens the request when that region changes, so a widening renewal isn't lost. Floors keep their exact whole-prefix meaning, there's no wire change, and the draft gains one SHOULD. A multi-region scope still certifies nothing. Best done as its own prerequisite PR, with this one rebased on it.
2. Floors carry regions on the wire (lite-07-wip). This reverses epic decision 10 ("a marker, not a region list").
3. Land this PR with the finish lanes accepting offline, and do 1 later. Not recommended: until then readers can't tell a finished broadcast from a lost one, which is the point of this quest.

Public API impact (breaking, no wire change)

• Rust: RouteEvent::Offline; announce::Event::{Unavailable, Available, Offline} beside Present/Absent; RouteInfo::available().
• JS: listing event kinds unavailable | available | offline, route kind offline, RouteInfo.available.
• FFI: MoqAnnounceEvent::{Unavailable, Available, Offline}, MoqRouteInfo.available. C ABI: MOQ_ANNOUNCE_KIND_{OFFLINE,UNAVAILABLE,AVAILABLE}, moq_route_info.available. Go, Python, Kotlin, Swift and Dart aliases and docs are mirrored.

Impact / migration

An exhaustive match on announce::Event or RouteEvent needs the new arms. To keep the old behavior, treat Offline like Absent/End and ignore Available/Unavailable.

Choices made without asking

• Kept the name Absent and did not rename it to Ended.
• Availability rides on Update/RouteInfo; the listing turns flips into Available/Unavailable.
• A session drop, an auth limit loss, or a narrower interest goes offline. moq-transport ends a route only on NAMESPACE_DONE or a clean stream close.
• An unavailable route ranks with draining routes.
• Overlap with moq-dev/moq#8369: whichever lands second turns RouteEvent::Offline into its route kind.

Suggested follow-ups

• The Waiting item above. Recommendation: option 1 as a prerequisite PR.

Checks

• just check: pass. Rust and JS unit and integration tests: pass.
• just test interop --all: fails in the finish lanes only, cause above.

Caused by

None: this is a feature.

Models

Size: M
Implementer: Claude Code, Opus 5.5, unknown effort

(Written by Opus 5.5)

🤖 Generated with Claude Code