Expand description
Axis 8: LLM-judge (user-supplied rubric).
This module defines the Judge trait that users implement (usually
in Python — see python/src/shadow/llm/). The Rust side only provides
the trait and the aggregation logic; no Rust-side default evaluator
is included, because calling an LLM from Rust is out of scope for v0.1
(SDKs are Python-first per CONTRIBUTING.md).
Traits§
- Judge
- User-supplied evaluator that scores a single (baseline, candidate)
response pair. Scores are in
[0.0, 1.0]where 1.0 means “candidate is at least as good as baseline.”