Case study · Self-learning

A strategy that learns — measured, not asserted.

Three specialist agents, one synthesising boss. Over a short backtest it held positive through the early dip, lagged the mid-window rally, then closed the gap as it accumulated decisions.

Relative capture, rising with experience

The signal is the trajectory, not the single number.

42%

First-half capture of SPY

78%

Second-half capture of SPY

109%

Final 10 sessions — +4.05% vs +3.70%

Its share of SPY's move climbs from under half to fully matching — capture rising with experience, exactly what a learning system should show.

The full window

Honest about absolute return.

~40 SESSIONS · BACKTEST

AG-3 +9.05%

Apr 5 – May 31, 2026, cumulative return.

SAME WINDOW

SPY +15.70%

AG-3 trailed in absolute terms over the full window.

DRAWDOWN

−2.0% vs −3.5%

It did so at roughly half the drawdown of the benchmark.

We show the lag, not hide it. AG-3 traded return for control and improved as it learned. AG-3 v2 and a six-agent AG-4 are in development.

How the loop learns →