The Production Model — Explorer

42,819 blind predictions, 40 refuges, 34 seasons. This is the app's own admin report card, rebuilt in public from the committed artifacts.

← Read the article

Grade strips — 34 seasons, graded like the app grades itself

one chip per blind season · click a chip to load that season below

Walk-forward: predicted vs actual

The trust table — every λ the backtest tuned

λ=1 full trust · λ=0 measured and ignored · hover a row for the story

When to trust it — per-condition report cards

out-of-sample skill inside each bucket of each factor