Priors

Both the compounded asset and the surface that builds it (formerly the Queue). As an asset: what the domain has compounded over time — established canon, character truths, world rules and lore, the structural shape of branches already explored, calibrated reads of how each actor moves under which pressure. The moat is your canon; its depth and freshness decide the result, the math is fixed and cheap. As a surface: a perspective-based prior-collection system. Members record observations against the world's open questions — filtering the noise, structuring each prior, setting the open threads worth tracking (web search assists the drafting; curation, not capture, never scraped). Priors are per-perspective — each seat keeps its own, a general perspective by default — and read out a running stance on how each thread bears as the story accumulates: a local bearing that moves over the chronology its priors were added. Divergence across seats is preserved (the antagonist's seat holds deliberately hostile priors; what's shared is the board, not the belief). A seat's running prior-set is a Stream (above); folding a seat's Priors into the domain commits its Stream via a Merge, and that committed stance guides scene generation — the next N scenes that carry the story forward. AI-simulated seats: because Priors are per-seat, an empty or quiet seat can have its priors simulated by the engine from its stances and open questions — keeping a thin-rostered room whole, putting AI agents in seats beside humans, and at the limit running a fully autonomous table. Humans are the main support; simulation lets the loop run when a human isn't there.