How to Track AI Trading Signal Accuracy With Real Outcomes
Learn how to measure AI signal performance without cherry-picking, inflated samples, or confusing confidence scores with win probability.
Save every eligible signal
Tracking only the calls that worked creates selection bias. Save the original thesis before the outcome is known and keep missed or cancelled setups visible without counting them as wins.
Separate resolved and unresolved ideas
A measured hit rate should use only signals explicitly marked Won or Lost. Waiting, tracking, missed and cancelled ideas belong in history but should not inflate the score.
Segment the evidence
Overall accuracy can hide important differences. Review results by pair, timeframe, direction and evidence grade to identify where the process is strongest or weakest.
Respect sample size
A handful of outcomes is not a reliable edge. Keep recording results consistently and avoid strong conclusions until the sample is large enough to be meaningful.