Network COGs

Independent multiplayer peers, authoritative requests, lost state and recovery controls.

Do COG requests and authoritative responses work across independent headless multiplayer peers?

Observe independent peers

This internal test level exercises stock-verb COG requests and authoritative responses across separate headless engines. Scenarios join a host and clients, add a late client, and replace a departing client. Local COG symbols and packet traces provide evidence of what each peer actually received.

The optional test transport lives in a separate tooling-enabled core build. An ordinary player build with that transport disabled was checked separately; the fixture is not a new player-facing networking mode.

Make failures observable

Ordered and reversed peer schedules check trigger counts, payloads and identity, local callbacks, object creation and movement, late joining, reconnection and input-driven respawn. Four negative controls drop triggers, corrupt a payload, misroute a unicast or mark authoritative creation as unsynchronized. Each must join successfully before failing its intended assertion.

A separate known-limit case reproduces a missing-resend defect after an early request is dropped. Its expected failure is evidence of a limitation, not successful packet-loss recovery.

Recover a model, not just an event

The existing DIGITRON protocol provides a second test: a real client button press changes the server, a late join receives a complete model, dropped live state creates measurable divergence, and RESYNC restores agreement without inventing an extra live-event cue.

Recorded result and boundaries

The 2026-10-11 gate completed eight rows, including the deliberately expected resend failure. API bounds and queue checks also passed. A separate ordinary-core smoke test exercised input recording, native state restoration and replay.

This page publishes the research findings, with no downloadable player level or Library release. Retail Jedi Knight and an attended multiplayer session have not verified the fixture. It does not claim that all packet-loss behavior is solved.

Back to Research

Independent peers make network behavior observable. The checks include negative controls and a known missing-resend failure; passing cases do not establish complete packet-loss recovery.