lpgcast.bench

The benchmark: a corpus of texts with the graphs they state, a scorer that needs no model as judge, and the lpgcast-bench command.

Modules:

Name	Description
cli	lpgcast-bench: cast the corpus and score it, or compare two runs.
compare	Paired comparison of two benchmark reports, episode by episode.
corpus	The benchmark corpus: texts, the schema they are cast against, and the graph
oracle	The oracle: a model that replies with the expected graph.
report	Rates from pooled counts, and the Markdown and JSON reports.
run	Run the corpus through a model, or the oracle, and score every episode.
score	Score a cast episode against its expected graph.