AI AGENT SWEEPS, WHEN SWITCHED ON · RUN IN-REPO BY THE REPLAY ENGINES · READ-ONLY

RESEARCH

When AI agents are switched on, they can request parameter sweeps; the same replay engines that produce every record on the cockpit run them in-repo. Every run lands here whole, losers included.

▸HOW TO USE THIS PAGE
  1. Each card is one run: the target (an indicator's parameters or an exit policy), the chart, the grid size, and when it ran.
  2. The table lists every variant tried with its measured record. Nothing is hidden to make the winner look better.
  3. BEST OF GRID is only named when a variant clears n≥30 — and it is survivorship-flattered by construction: the best of 200 tries is expected to look good.
  4. Promotion into the score or a setup machine never happens here. It needs the ADR-0014 gate: a replayed record, a calibration check, a written decision.
  5. Runs arrive from AI agents once they are switched on (≤200 variants a run, ≤4 runs a day). Until then this page stays honestly empty.

CAN'T TELL YOUThat a parameter will keep working. A sweep measures the past under one cost model; the ledger and the calibration panel measure the present.

LOADING…