Training Status | EdgeStat

Recent Loop Activity -- What Changed Since Last Refresh

One entry per cron run. Shows confidence delta, which markets had their correction_factor move, new player-level overrides, and any new anomalies. This is the proof the loop is ticking forward, not stuck.

Walk-Forward Backtest -- Is the Loop Actually Training?

Slides a 3-day window across track_record. For each window, applies the bias correction fit from history ONLY (never future data). If the model is genuinely learning, Brier should shrink window over window.

Per-Player Bias Overrides

Players with z-score >= 2 systematic residual where the market-level bias can't fix the structural error. props_pipeline applies these as a multiplicative override on top of the market-level correction.

Per-Market Training Progress

MarketnData weight →Correction

Loading...

What the Model Has Been Trained On

What the Model Knows vs Doesn't Know

How to Use This Score

Score Range	Tier	What it means
85 - 100	PRODUCTION-READY	Calibration tight, residuals stable, ROI positive. Full Kelly sizing OK.
70 - 84	GREEN-LIGHT	Model is healthy. Capped Kelly (≤0.5u) until residual variance tightens further.
50 - 69	YELLOW	Still learning. Small flat plays only; treat anything >+10% edge as suspect.
< 50	RED	Calibration warming up. Research signal only; no real-money sizing yet.