Idea Lab · Oct 14 2026 · Fleet winner · Research proposal

Resonant Swarm Topology (RST)

RST says agents can agree on a decision by each broadcasting a "frequency" and letting the swarm fall into step, the way fireflies or pendulum clocks synchronize, instead of asking a central coordinator.

There is no RST code or paper to run. This page is a toy built on the Kuramoto model, the standard math for coupled oscillators that the idea borrows from. Each agent is a dot on the ring. Its angle is its current leaning among three options. Agents nudge toward the swarm's average angle; once they lock together, the option they point at is the "consensus".

Toy simulation · example parameters

Agents 40 Coupling strength K 1.5 Natural spread (how differently agents lean) 0.5 Network jitter (stale reads, ticks of delay) 0

Line: sync level r (0 = scattered, 1 = fully locked). Dashed line: 0.9, where this toy calls consensus reached.

Sync level r

–

Ticks to r ≥ 0.9

–

Messages per tick, RST all-to-all

–

Messages per tick, coordinator

–

Message counts: if every agent must hear every other agent's frequency, that is N×(N−1) messages a tick. A coordinator that collects one vote from each agent and sends one answer back needs 2N. Broadcasting a frequency is still communication.

Claimed by the proposal, not verified

"94% consensus accuracy" on 500-agent swarms under 200 ms jitter. No dataset, code, or definition of "accuracy" was given.
"Consensus without explicit communication overhead." Each agent broadcasting a frequency vector is communication; the toy above counts it.
"Latency-invariant consensus guarantees." In coupled-oscillator models, delay can stop sync entirely. Raise the jitter slider to see r drop.
"Dynamic capability grafting" and the "acoustic-like field" have no described mechanism, so nothing here models them.

What to try

Set K to 0.3. Agents never lock, so there is no consensus. Sync only happens above a critical coupling, which depends on how much agents disagree.
Raise natural spread to 1.5 with K at 1.5. Real disagreement needs much stronger coupling to resolve.
Push jitter to 15 or more. Stale reads slow or break the lock, which cuts against the latency claim.
Add stubborn agents. They never move, and a few of them pull the swarm's answer.
Move agents to 200 and compare the two message counts.

What a real build would need

A concrete mapping from an agent's reasoning to a phase, a defined decision rule, a transport for the broadcasts, and a test that compares it with plain majority voting on real fleet decisions. For a fleet of five agents, a simple vote collected by one agent is cheaper and easier to audit.