Server and core: version ranges on both sides with a coded refusal; a
client id maps to the same member and token for the server's life; a
drop is `gone` for a 10-minute grace keeping ready, presence, queue
items, relay and avatar; `bye` leaves at once; ordered takeover (seq,
replaced/stale); presence joining/in/away/gone with one PresencePolicy;
joiners never pause playback and hold a held gate at most ~15 s;
state.hold; feed ids, seq and replay; the relay lives through its
source's grace and continues on re-dial; queue `by` lasting, auto
items and cancel; MediaRef carried; client ids never logged.
App: persisted client id and seq; Connection with unstable/lost/
replaced states, backoff with jitter, no lost/back pair per flap; a drop
keeps the session (state, player, media, relay source); presence in the
strip and members card; sit out = away intent; feed matched by id;
auto sit-out (idle when not playing, minimised for a while; never when
alone) replaces auto-unready; bye on close and Leave, not on update
restart; relay source re-dials through its grace.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- notice_table_test: one row per AppEvent variant through noticeFor.
- FlakyProxy (core/testing.dart): pass, blackhole (half-open), delay,
hold, cut after N bytes, reset; demo tests against today's server.
- Liveness (stale/drop/ping) on the engine clock, so grace can be tested
with a fake clock; Connection takes an injectable clock.
- AppDirs with STARSYNC_PROFILE=<name>: separate data and logs, so two
instances run on one machine.
- DEVELOPMENT: scenario list per restructure step; test tags follow the
new step order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>