Complete portable test suite covering the LiveObjects path-based API:
21 files across unit tests (pure + mock WebSocket), integration tests
(sandbox), and proxy integration tests. Covers PathObject, Instance,
BatchContext, LiveCounter/LiveMap CRDTs, ObjectsPool sync state machine,
value types, subscriptions, and GC. Includes table-driven validation
tests, bytes/binary data coverage, and REST fixture provisioning.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

The integration GC tests (RTO10, RTLM19) use enable_fake_timers() +
ADVANCE_TIME(), which only affect the SDK's local clock, not the sandbox
server. The same scenarios are already covered by unit tests at
realtime_object.md (RTO10) and live_map.md (RTLM19) where fake timers
work correctly against the mock WebSocket.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Data-path integration tests must run with both JSON and msgpack per G1.
Add ## Protocol Variants section and useBinaryProtocol to ClientOptions
in objects_lifecycle_test.md and objects_sync_test.md. Update the
annotated specs list in integration-testing.md.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

root is the PathObject at / (a LiveMap). Calling increment/decrement on
it would throw 92007 (RTPO17e). The tests intend to operate on the
counter at root["score"]. Fixed in realtime_object.md (10 sites) and
live_counter_api.md (5 sites). path_object_mutations.md is unchanged —
those are intentional negative-path tests for RTPO17e.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Tests use instance.subscribe() (RTINS16) which emits
InstanceSubscriptionEvent { object, message }, not LiveObjectUpdate
{ tombstone, objectMessage }. Fix all assertions to use .message
instead of .objectMessage, and check .message.operation.action instead
of .tombstone. Also fix noop test serial from "01" to "02" so the
newness check (RTLO4a6) doesn't suppress it before the noop path fires.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Per RTO7a/RTO7b, these attributes belong on RealtimeObject, not
ObjectsPool. Add `realtime_object = RealtimeObject(pool: pool)` to
affected test setups and redirect all references (8 tests, ~11 lines).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace non-existent getRoot() with get(), and restructure so root is
obtained while attached then channel is detached before subscribe().
The previous setup responded to ATTACH with DETACHED, so get() would
throw 90001 before reaching subscribe().

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Replace private _liveObject.siteTimeserials access with an observable
test: after a local increment (applied via ACK with source LOCAL),
send an inbound COUNTER_INC with the same siteCode and serial. If
LOCAL incorrectly wrote to siteTimeserials, the newness check would
reject the inbound message. The counter reaching 120 proves LOCAL
did not pollute siteTimeserials.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Instead of just checking score == 100 after re-sync (which proves nothing
since OBJECT_SYNC resets the pool regardless), replay a previously-applied
serial after re-sync. If appliedOnAckSerials was properly cleared, the
replayed message applies normally and score reaches 110. If not cleared,
dedup (RTO9a3) rejects it.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Add missing protocol message actions from TR2: OBJECT (19),
OBJECT_SYNC (20), and ANNOTATION (21).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>