promptdojo_

Retraining triggers and human review — step 7 of 7

Checkpoint

One last thing before we move on. Same surface as a write step — but the lesson doesn't complete until this passes.

Run the full policy over a month. The POLICY dict carries thresholds WITH owner and date — print its header line first: f"policy: {d}d schedule, {floor} floor (owner {o}, set {date})". Then decide() per week: the upstream-bug veto returns "VETO - fix data, no retrain"; otherwise collect schedule/drift/outcome reasons into "RETRAIN a+b" or "hold". Print f"{name}: {decision}" for W1-W4. Note W2: drifted AND bugged — the veto wins.

full-screen editor opens — close anytime to keep reading.