Reveal Mode controls round reveal semantics for the seats, not your own operator visibility.

Blind: the round stays blind externally until completion. Seats can still build internally if Turn Policy is Chained.

Free: normal open round behavior.

Wait-for-all: the round waits for everyone before it is treated as complete.

Turn Policy controls how the LLMs run inside the round.

Parallel: every seat answers independently.

Ordered: seats run one at a time, but each gets the same prompt.

Chained: seats run one at a time, and later seats see earlier same-round answers.

Chained Passes only matters when Turn Policy is Chained.

1 pass: one telephone walk through the lineup.

2+ passes: loop through the same seats again in the same round.

Governance Mode controls the system prompt weight.

Full: infrastructure/ops mode. Seats get master docs, runtime injection (plugins, fleet state, MCP tools), and the 6-step compliance format mandate. Use for audits, ops tasks, code review.

Lightweight: task-only mode. Seats get identity + room instructions + the task. No compliance format, no runtime injection, no master docs. Use for games, stories, brainstorming, Q&A.

Examples: Blind + Chained = seats build on each other, but the round stays blind until complete. Free + Parallel = seats answer independently. Lightweight + Chained = telephone game with no compliance noise.