The runs
Every recorded run, with the configuration that distinguishes it and the scorer's decision on it.
These are the real runs against live API calls. They were not all
configured alike: the tool surface changed between versions, the per-vintage
coverage scopes were withheld in one, the data dictionary's populated
field was suppressed in others, and the turn and call ceilings moved. A run the
scorer refused is one that stopped before the agent said it had finished; a
partial answer is not scored.
| Run | Canary | Established by | Tools | Scopes | Retrieval | Caching | Termination | Scorer | Turns | Calls |
|---|---|---|---|---|---|---|---|---|---|---|
Loading the run index… | ||||||||||
How the canary's presence was established
The wording below is the export's own, unedited. The short name in the table points at one of these.