Out-of-sample · No data leakage
Performance Analytics
2-year out-of-sample backtest · $10k starting capital · KPI cards match Portfolio Replay
🧠 20-Agent AI Council12/17 weighted votes
Year-by-Year Performancesame strategy, different windows
Loading year stats…
Historical Win RateLoading…
≥55% — strong edge
45–55% — moderate
<45% — weak
Win rate = % of signals where the 5-day price move matched the predicted directionBuilding performance cache — takes ~2 min on first load
Out-of-sample results — signals evaluated on data the model never trained on. Backtests run on today's ticker universe, so absolute returns carry survivorship bias and read optimistically; the per-signal edge vs the universe average (and the live weekly track record) are the more reliable measures. Past performance does not guarantee future results.
Factor AttributionProprietary scoring — contribution of each market factor to today's signals
Loading SHAP analysis…
All results are out-of-sample — signals generated strictly on unseen data. Past performance does not guarantee future results. Not financial advice.