Can Matrix carry a tactical radio network?

Live demonstration

Replay of a real run. Two convoys (blue / orange); the link between them flaps. Watch cross-convoy messages stall during a partition and burst through on the heal — store-carry-forward in action.

First results

Every figure below is generated from real orchestrator logs of the pilot campaign (see methods). Headline numbers:

Per-run summary

Scaling to N=70: one requirement, one real barrier

⚠ Results updating. An earlier headline ("a fundamental convergence ceiling, ~37% at N=70") was traced to a setup artifact, not a protocol limit. The large-N scaling is being re-measured with the corrected setup (ETA a few hours); figures below refresh automatically.

A setup requirement — pre-establish the room (not a hard limit)

Naively joining 70 servers concurrently rate-limits the resident server and forks the room DAG, so membership never converges and delivery looked capped (~33%). But this is a deployment procedure, not a barrier: with sequential pre-establishment + a convergence gate (form the operational room while connected, before dispersing), all 70 servers converge — and a converged N=70 room then delivers ~95% at ample bandwidth.

The real barrier — bandwidth / fan-out

Each message fans out to N−1 point-to-point federation transactions of ~762 B each, through one shared radio. So required per-radio bandwidth grows with fleet size (the fan-out tax): the fluid law r·(N−1)·(S+O) sets the shape, and real throughput sits well above it (federation is round-trip/chatter-bound). This is what makes a 70-vehicle operational room infeasible at tactical link rates.

How it works

Only the server-to-server (federation) links are shaped; the client↔own-server link inside each vehicle is never impaired. See the paper for the full method, the closed-mesh gotchas, and limitations.