Zucca Evals

Extraction accuracy across ingredient & supplier fields.

Whether two ingredient or supplier records are the same thing: the LLM merge judge scored against hand-labelled pairs, with false merges as the headline.

Explore ingredient and supplier records as projections and similarity graphs. Labels and predictions are optional overlays.

Hybrid Smart Specs search: ranking quality on a frozen real corpus.

How often hybrid search returns the spec a formula was historically bound to — agreement with the old matcher, not accuracy.

Prod and staging model spend by tool, model, environment, user, org and product. Zucca staff only.