attempt ledger
A structured log that records each run attempt of an AI agent, typically implemented as an append-only JSON Lines file. It records the model name, verdict results, elapsed time, token counts, and the decision basis for failure gates, facilitating later statistics and attribution.
Timeline (last 90 days)
Without an attempt ledger, developers cannot distinguish between "the model is too weak" and "the harness is broken," nor compare two models without guessing.
The key constraint of the attempt ledger is to write immediately after each attempt, not to batch-record at the end of the day; a reject must also carry a gate name (one of build, scope, or proof).
All Facts (2)
Without an attempt ledger, developers cannot distinguish between "the model is too weak" and "the harness is broken," nor compare two models without guessing.
50%UnverifiedThe key constraint of the attempt ledger is to write immediately after each attempt, not to batch-record at the end of the day; a reject must also carry a gate name (one of build, scope, or proof).
50%