Module eval 

probe eval: the user’s prompts against one or more models, deterministic sampling.

A prompts file is JSONL; each line is an object with messages (or prompt) and any of tools, expect_tool, expect_args (a subset the arguments must contain), schema (sent as response_format and checked), expect_contains, and an id.

Structs

Row

RowResult

Functions

parse_jsonl

run

score

subset

want is contained in have: objects by key recursively, everything else by equality.