Module benchmarks 

Capability-benchmark registry (CH-09 capability-registry-benchmarks).

Extends the compile-time model catalog (crate::llm::catalog) with independently-sourced benchmark scores, so the super::router::ModelRouter can break cost/context ties using measured capability instead of catalog order alone.

The dataset in src/llm/data/model_benchmarks.json is hand-curated, not scraped — see src/llm/data/UPDATE.md for why and how to refresh it. Every entry MUST carry a source_url and retrieved_date per docs/uar-next-fable.md §2.4 (“version every entry with a source URL + retrieval date”).

Structs

BenchmarkScore

A single sourced benchmark measurement for one model.

Enums

BenchmarkDimension

Which capability dimension a benchmark measures.

Functions

best_score

Best known score for a provider/model id on the given dimension, if any.

scores_for

Look up all sourced benchmark scores for a provider/model id.