use-ai-eval

AI evaluation metadata primitives for RustUse.

Experimental

use-ai-eval is experimental while use-ai remains below 0.3.0.

Example

use use_ai_eval::{AiEvalKind, AiEvalRunId, AiEvalScore};

let run = AiEvalRunId::new("eval-001")?;
let score = AiEvalScore::new(0.92)?;

assert_eq!(run.as_str(), "eval-001");
assert_eq!(score.value(), 0.92);
assert_eq!("tool use eval".parse::<AiEvalKind>()?, AiEvalKind::ToolUseEval);
# Ok::<(), use_ai_eval::AiEvalError>(())

Scope

• Eval run identifiers, rubric names, bounded scores, eval kinds, targets, judges, metrics, datasets, outcomes, and failure modes.
• Metadata only.

Non-goals

• Running evals, judging outputs, calling models, scoring datasets, collecting latency, or calculating costs.

License

Licensed under either Apache-2.0 or MIT.