Recover dropped WebRTC routes without replacing the player seat

[sbddesign]

When a connected player's WebRTC route drops, try an ICE restart on the existing connection so their data channel and seat survive. Either side can notice the outage; the joiner owns restart offers. If recovery fails within the existing eight-second grace period, retain CONNECTION LOST and manual RETRY.

Keep signalling alive until channel close. Bind restart messages to the established peer and attempt number; buffer early candidates, retry lost offers/answers, and remove recovery listeners on leave or failure. No new infrastructure or gameplay protocol version is required.

Validation:

• Full typecheck, 737 simulation assertions, balance, and production build pass.
• Chrome/WebRTC: new ICE credentials with the same peer connection, data channel, seat and advancing match clock; recovery despite two seconds of dropped answers; host-triggered recovery; blocked offers lead to RETRY. Zero page errors.
• GitHub full mutation gate: 147/147 caught cleanly, zero survivors or untestable cases. All CI checks pass at cf6b284 (run).

The browser check injects an outage notification and exercises real renegotiation over Nostr. A physical Wi-Fi interruption across two devices remains unverified. Recovery cannot replace TURN when no direct route exists or keep a match alive after its host leaves.

The local gate also printed 147 clean catches and restored the worktree, but its supervisor hit the ten-minute cap; the successful complete-run verdict above is from GitHub.

Preview: https://neon-orbit-m5abgrh1p-delineatormedia.vercel.app/?host=2

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
neon-orbit	Ready	Preview	Sep 6, 2026 8:33pm UTC