bayes replay: in-play policies on synthesized live markets

← lab · model run 18 · in play vig 0.06 · matches replayed 3,800 · ran 7/7/2026

What this is: A one-off study — its numbers below are exactly what the run reported.

note	Live market is SYNTHETIC (closing-line lambdas conditioned on the live state) with realistic book shading: base vig + tail penalty growing with price, quotes capped at 34. roi_fair grades at unshaded Poisson-fair prices (upper bound / diagnostic only — real books fatten comeback tails). Overreaction edges are invisible here by construction; grading vs REAL live odds accumulates in af_live_odds.
model run	18
in play vig	0.06
matches replayed	3,800

policies

bets	policy	staked	seasons	hit rate	roi fair	roi at vig
1,324	fade_overreaction	13,240	10	0.167	0.5105	0.1168
4,063	double_down	40,630	10	0.294	0.1444	0.0572
3,216	prematch_only	32,160	10	0.329	0.0112	0.0112
5,042	cover_lock	33,584	10	0.252	0.0161	0.008
3,789	value_stream	37,890	10	0.23	0.2646	-0.022

AUC	predictability: 0.50 = coin flip, ~0.70 = ceiling for sports
Importance	how much the model leans on this factor (permutation importance)
Direction	sign of the raw correlation with the outcome
Survives all eras	effect points the same way in every historical era

Spec · the reproducible recipe

{
  "kind": "bayes_replay",
  "stake": 10,
  "live_edge": 0.08,
  "model_run": 18,
  "in_play_vig": 0.06
}