Observed arrival · 2026-10-15
Evidence Deflator: Can a trading record beat luck?
An open statistical tool assesses whether trading records—especially Hyperliquid wallets and AI backtests—look distinguishable from luck after accounting for selection and training-data exposure.
- For
- Reviewers of trading records and AI backtests
- Worth noticing
- A worked example marks 58 of 66 months as potentially seen before a model's cutoff, leaving eight clean months against 48 needed at Sharpe 1.
Field notes
The method divides a test window into months that may overlap a model's training data and months that count as clean evidence. It also raises the threshold when a result is picked from many attempts: the page gives t ≥ 3.26 for the best of 41 and t ≥ 4.76 for the best of 23,376. A Python command is shown for reproducing the worked window example, and the homepage links to Python, CLI, MCP, and API options.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue