Every strategy idea — generated, variation-swept, and tested against 12 months of CME tick data — lives in this corpus. Walk-forward validation, regime decomposition, full kill log. The survival rate is intentionally low; only the strategies that clear the bar are promoted to production.
A strategy is just a hypothesis until it survives all three. Most don't. That's the point.
Run against 12 months of 1-min CME data. Compute PF, win rate, mean edge, max drawdown. PF floor: 1.2. Sample size floor: 30 signals.
Train on the first window, test on a held-out trailing window. If OOS PF doesn't hold within 20% of full-period PF, the strategy is killed regardless of in-sample edge.
Performance is decomposed across vol regimes (low/normal/high) and day types (trend / balance / reversal). A strategy that only works in one regime gets flagged for conditional production.
A read of the corpus shape: outcome by grade, category mix, and edge distribution among profitable strategies.
Every row is a hypothesis we tested and graded.
| Date | Name | Category | Grade | N | WR | PF |
|---|---|---|---|---|---|---|
| Loading manifest… | ||||||
Different cuts on the same corpus.
Every tested strategy sorted by PF, μ, and survival status.
Filter the corpus by grade, category, and depth.
Daily pivot tag-rate timeline. Real production data.
Strategies running in production. Live performance audit.