adk_eval_ensemble (erlang_adk v0.10.0)

View Source

Deterministic judge ensembles and threshold calibration.

This module aggregates only already-bounded judge scores. It deliberately contains no provider lookup or model invocation, making an ensemble result reproducible from a persisted set of votes.

Summary

Functions

aggregate(Votes0, Options0)
apply_calibration(Score, Calibration)
calibrate_threshold(Examples0, Options)

Functions

aggregate(Votes0, Options0)

-spec aggregate([map()], map()) -> {ok, map()} | {error, term()}.

apply_calibration(Score, Calibration)

-spec apply_calibration(number(), map()) -> {ok, map()} | {error, term()}.

calibrate_threshold(Examples0, Options)

-spec calibrate_threshold([map()], map()) -> {ok, map()} | {error, term()}.