StockPicker Backtest

Overall

Cumulative Return

Running sum of realized per-pick returns in run-date order. Assumes one equal-sized position per pick; not compounded. The slate line is the same running sum of the index-ETF return over each pick's holding span — the gap between the lines is cumulative alpha (picks without a captured benchmark advance the slate line by 0).

Success Rate by Time Window

Hit rate covers resolved picks only (including early wins still inside their holding window); Pending shows how many picks from that window haven't resolved yet.

Rolling Win Rate

Hit rate over the trailing 10 resolved picks whose holding window has fully elapsed, plotted at each pick's run date. Excludes picks that hit their target before the window closed (which would bias the recent tail to 100%), so the rightmost point trails today by roughly the holding-period length. Reference line at 50%.

Per-agent accuracy

All-time hit rate and average return across every resolved pick each agent backed. "Solo" counts picks that agent backed alone. The ensemble weight is what actually scales that agent's vote in ranking — its hit rate over the trailing 40 backed-and-resolved picks, shrunk toward 1.0 below 20 samples (weight > 1 = sharper than a coin flip). Weight columns show "—" until weight data exists.

Performance by pick position

Probability Calibration

Compares what the agents predicted against what actually happened. Calibrated rates use isotonic regression to enforce monotonicity.

Top 5 winners / losers

All Picks