Module eval 

judge() — one-shot quality scoring against a goal + criteria.

Rust port of node/src/runtime/eval.ts. Wraps the kernel’s gen_eval free functions (re-exported from deepstrike_core::harness::eval) into a small surface for callers that just want “does this result meet the criteria?” without standing up HarnessLoop.

Structs

Criterion

Re-export the kernel’s Criterion so callers don’t need to import from deepstrike_core. A single evaluation criterion with optional weight and required flag.

Functions

build_eval_messages

Build the kernel’s eval prompt for (goal, criteria, result). Pure — does not call an LLM.

judge

One-shot judge: render the eval prompt, stream the provider, parse the verdict. Errors when the provider returns no text.

parse_verdict

Parse a Verdict from raw judge-LLM text.

verdict_output_schema

The JSON Schema the kernel expects judge output to conform to.

Type Aliases

Verdict

The verdict shape parsed from the judge LLM. Same struct as the kernel’s EvalResult.