adk_eval_case_metric behaviour (erlang_adk v0.7.0)

View Source

Full-case evaluator behaviour.

Unlike the legacy per-turn adk_eval_metric callback, this behaviour sees all actual invocations, the canonical trajectory, and the expected case.

Summary

Types

result/0

Callbacks

score_case/2

Types

result/0

-type result() ::
          number() | {ok, number()} | {ok, number(), map()} | {not_evaluated, map()} | {error, term()}.

Callbacks

score_case/2

-callback score_case(EvalInput :: map(), Config :: map()) -> result().