Long-form on the
research process.
Field notes from the bench. The ML pivots, the acceptance-gate stories, the failed iterations we publish anyway because the methodology depends on you seeing them.Field notes
Iter 20: the dead-zone filter that lifted CAGR 0.78pp
We found a uniquely negative bucket in the GBDT signal distribution (1.80-2.00, 39.9% win rate). Filtering it lifted wf-CAGR +0.78pp and dropped stitched DD -2.49pp. Free wins like this are rare; documenting how we caught it.
May 9, 2026
ReadWhy hard-blocking weak ML signals hurts more than it helps
Phase X.3 → AA pivot story. We tested binary block-on-low-prob and watched OOS Sharpe degrade by 0.18. Soft-sizing the same model preserves trade volume and the lift survives — here's why.
Apr 22, 2026
ReadHow to build a soft-sizing entry gate from a logistic regression in 200 lines
Annotated walkthrough of vectra-ml's entry classifier. 10 features, 6.4k training samples, sklearn-style fit/predict, exported to JSON for the live engine. No frameworks.
Apr 8, 2026
ReadRealistic crypto-bot Sharpe expectations: a 4-year audit
What sustainable crypto-momentum Sharpe actually looks like across regimes. Includes the 2022 bear, the 2024 ETF run, and 2025-2026 chop — folded out per period so the marketing-screenshot bias goes away.
Mar 19, 2026
ReadMore posts land as research lands. Subscribe to the waitlist on the home page and get a ping when each one ships.