[Taure]

asobi v0.84.1 (widgrensit/asobi#479) made world.ack monotonic per connection. asobi_player_session holds a last_ack_seq and drops any ack that does not advance it; the session is per WebSocket and dies with it, so the mark resets exactly when the client's counter does.

The Unity docs told readers to keep a running maximum and ignore any ack that did not advance it, because until v0.84.1 the mark was recorded per zone and a crossing player could receive a lower seq from the zone they left. That can no longer happen, so the workaround now defends against something the server guarantees against.

Removed

• The per-zone paragraph: the mark being "held per zone, not per connection", zones acking independently, more than one ack per broadcast after a crossing, payload.seq going backwards, and the instruction to keep a running maximum (README.md).
• The widgrensit/asobi#477 reference. The issue is closed.
• The long _acked = -1 high-water variable and its if (ack.seq <= _acked) return; filter from the reconciliation sample.
• The claim that acks keep arriving with an older mark after an out-of-range seq. The monotonic filter drops a non-advancing ack, so they do not.
• The "nine world.ack frames per broadcast" half of the broadcast_interval bullet.

Reconciliation is now simply: drop every buffered input at or below ack.seq, then replay the rest on top of the accumulated state.

Version floor

The README bullet is a bare floor: Requires asobi core v0.84.1 or later. Nothing about what an older server does, because v0.84.0 does send acks; they can simply regress across a zone crossing.

The AsobiDispatcher comment stays at v0.84.0. It records when world.ack first appeared on the wire, which is provenance, not the floor for the monotonicity guarantee.

Reserved data key

Narrowed to its v0.84.1 rule wherever the docs mentioned it: only a payload whose sole key is data mapped to an object is unwrapped (deprecated, removed at the next protocol break). A data alongside other keys, or one whose value is not an object, is forwarded verbatim - siblings are no longer dropped. A payload that is not an object at all draws an error frame with reason invalid_payload rather than being read as empty input, and WorldInputAsync's XML docs now tell that same story as WsFrame instead of claiming the server cannot report a bad input at all.

Kept, unchanged

The opt-in rule, the high-water-mark semantics, a rejected input still advancing the mark, the ack being addressed to one connection and never riding the shared world.tick, the world.tick delta/accumulate guidance with its a/u/r ops, the full-snapshot-per-new-subscription and interest-ring behaviour, and broadcast_interval defaulting to 3 with 1 giving an ack every tick.

Docs and comments only - no behaviour change. dotnet test Tests/AsobiCore.NET: 128 passed.