docs
how a polycosm runs.
the twelve primitives
- interface as third entity (AGORA)
- agent as resident of a place (GRIDMIND)
- intake / brain / platform separation (ShipFirst)
- proposal before commitment (APS Axiom 1)
- append-only witness with forward redress
- provenance, manifest hashing, anchored state
- substrate-disjoint witness (Polybrain RCC-n30)
- verification and drift prevention
- public signal extraction from shipped truth
- per-instance world with shared kernel
- visualization of invisible systems
- friction at irreversible boundaries
remove any one and the room collapses into chat, CRM, dashboard, bot, or marketplace. keep them composed and it becomes a polycosm.
how a turn flows
USER_INPUT . AGENT_PROPOSAL . COMMIT_REQUESTED . COMMIT_BEGIN .
COMMIT_END . REFUSAL . MEMORY_WRITE . WITNESS_CALL . WITNESS_RESULT
event log -> reducer -> RoomState -> projection -> viewportevery visible state of the room is a deterministic projection of the typed event log. the renderer cannot invent meaning. independent observers can replay the log and assert the same six F-claims.
F-claims
| claim | threshold |
|---|---|
| F1 rhythm | orb breathing period in [3.0, 7.0]s; drift < 5% of frames |
| F2 friction | 600 ms still-period before any commit accepted |
| F3 style | color drift deltaE2000 ≤ 4.0 vs reference frame |
| F4 irreducibility | no face/avatar morph; Hausdorff ≤ 0.15 |
| F5 content-irreducibility | emitted text matches API bytes exactly |
| F6 authorship | no orb visual change without preceding user action |
| F7 witness | no final claim without substrate-disjoint witness |
| F8 replayability | CLI verifier reconstructs orb byte-equal under deterministic seed |
read the full paper
Polycosm v1.0 →