claude/probability-config-review-1tdolw #119
Loading…
Add table
Reference in a new issue
No description provided.
Delete branch "claude/probability-config-review-1tdolw"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Two problems addressed: 1. Favorites' P(1st) was too sharp (e.g. NHL top teams ~20% vs ~12% implied). - The NHL simulator hardcoded its parity factor (1000) and ignored the season config's parityFactor, so the knob meant to flatten the distribution did nothing. It also re-blended raw futures odds into every game on top of the odds->Elo conversion, double-counting the same signal. - NHL now reads parityFactor/iterations/seasonGames/overtimeRate from config and no longer re-blends odds per game (odds enter once, via the central odds->Elo resolver). Honoring parity 2500 flattens a top team from ~29% to ~13% title odds. 2. "Season Config" and "Input Policy" were two forms over the same stored object that didn't reflect each other, and the engine-knob half was inert for many simulators. - Every simulator now reads its engine knobs (iterations everywhere; parityFactor for all Elo-based sims) from the merged config, passed in by the runner via the Simulator interface. Defaults equal the former hardcoded constants, so behavior is unchanged unless a season overrides. - The admin simulator page is now a single "Simulator Configuration" card with structured Engine and Input-derivation sections (profile-driven, so each sport shows only the knobs it honors) plus an Advanced raw-JSON escape hatch — all writing the same config. Also: centralized the duplicated configNumber helpers into config-access.ts; the central odds->Elo resolver now maps onto the configured Elo floor/ceiling so those bounds set the odds-derived spread (a real flattening dial). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01PAFMogMkFJf52YpHyCDvuf