The Backtest Lie Detector is free, runs in your browser, and nothing you paste is uploaded. Here is exactly what it checks, with the numbers from two real runs of our own certifier's demo books.
Book 1 — the overfit one. The claimed accuracy was 78%. Recomputed from the prediction file: 69.33% (416 of 600). The five chronological folds told the story: 80.00% in the first fold, 55.83% in the last. The edge decayed as time moved forward — the classic signature of a rule tuned on the past. Verdict: OVERFIT, exit code 1.
Book 2 — the good one. Recomputed accuracy 83.67%. The folds held flat. Verdict: EDGE DEMONSTRATED, exit code 0.
Three checks, in order:
- Recompute it. The tool rebuilds the accuracy from your prediction file — date, prediction, actual, 1 or 0 — instead of trusting the number on the slide.
- Beat a dumb baseline. It scores always-bullish and no-change on the same window. If your strategy cannot beat doing nothing clever, there is no edge to discuss.
- Walk it forward. Five chronological folds, plus a Wilson 95% interval on the win rate. A 70% win rate over 20 trades is 48–85%; over 100 trades it is 60–78%. Same win rate, different truth.
The verdicts, in order: NO EDGE (accuracy at or below the best baseline), OVERFIT (last fold more than 20 points below the first), WEAK (the Wilson lower bound at or below the best baseline), EDGE DEMONSTRATED.
Try it on any backtest you are being sold, or your own: https://startbuildit.online/lie-detector/
A claim you can't check isn't a claim. It's an ad.
Top comments (0)