No user-submitted related servers found.
Scored across 6 tools
The boundaries are mostly clear: `evaluate` handles multiple questions over one state, `evaluate_batch` handles many items, and `noul`/`choice`/`score` are single-question shortcuts per question type. However, `noul`, `choice`, and `score` are strict subsets of `evaluate`'s capability, so an agent could justifiably always use `evaluate` and ignore the other three; the descriptions acknowledge this overlap but don't fully eliminate the redundancy.
All names are namespaced under `jev.` and use consistent snake_case (`evaluate_batch` matches `evaluate`). The only deviation is that `evaluate`/`evaluate_batch` are verbs while `models`, `noul`, `choice`, and `score` are nouns naming question types, which is readable but not a single uniform verb_noun pattern.
Six tools is well-scoped for an LLM-judge service: one general evaluate, one batch, one discovery tool, and three typed convenience wrappers. Every tool maps to a distinct use case and none feels like filler.
The surface covers the full evaluation lifecycle: single state, batch, model discovery, and all three question types (yes/no, choice, score), including image and structured-state support. Minor gaps exist—no tool to inspect the configured provider/model default, check credit or usage status ahead of a run, or resume partially failed batches.