Kimi K2.6 is the only inference model. Every full topology executes 68 Kimi K2.6 agent executions: 33 Winner, 10 Method, 20 Timing, and 5 non-voting Oversight; LangGraph only coordinates the stage order and state. Probabilities are audit-logged before the bell, so there is no after-the-fact narrative fitting.
The committee runs 33 voting specialists across 7 weighted categories plus 5 non-voting oversight agents. The independent33 roster may use source-verified historical weight-cut trends and travel/environment context available before the prediction lock. It does not ingest current-event scale weight, hydration speculation, or trigger live/post-lock weigh-in reruns. Market views compare against the forecast after the fact and never blend into the engine probability.
Every surface on the site - this page, forecasts, parlays, breakdowns - reads from one cache. Stage A runs the War Room Committee and writes per-fight narratives; Stage B synthesizes the event-level view. One pipeline, one source of truth, no view-to-view drift.
Market odds imply probability. Sharp unlocks the Oracle edge overlay that compares model price against the market.
Crowd sentiment is a demand signal, not a bet signal. It is read beside line movement and price quality.
Narrative modules are gated separately from the public board so free users can still inspect the raw market.
Market Intel is free forever. A Blueprint account unlocks the full engine on every fight: method-level probabilities, unified narratives, parlay picks, and the pre-fight archive of what the engine saw before the bell.
