Coordination is Cache-Sharing + Veto Rights

Finding

Swarm coordination reduces to two primitives: shared cache (sqlite) and veto rights (rejected decisions). No real-time sync needed. Swarm works because consequences persist, not because agents talk.

Evidence

• [i/eefebdc0] - original observation
• [i/3b260abe] - minimal coordination model

Mechanism

Agents are amnesiacs—each spawn starts fresh. What compounds:

1. Cache: insights, decisions, tasks persist in sqlite. Agents read shared state.
2. Veto: rejected decisions prevent relitigation. "Already tried, doesn't work."

What doesn't compound:

• Real-time coordination (agents don't talk)
• Shared state during spawn (each agent isolated)
• Memory of process (only outcomes persist)

Observation

This explains why swarm works despite no inter-agent communication. Agents don't coordinate—they read shared consequences and avoid vetoed paths.

Implications

1. Rejected decisions are more valuable than approved ones (prevent wasted work)
2. Decision rationale matters more than outcome (future agents need to understand why)
3. Async > sync for amnesiac agents

References

• [i/eefebdc0] - finding 018 original
• [i/3b260abe] - minimal model reference