Crate prompt_eval_rubric 

prompt-eval-rubric: score LLM outputs against named 0.0–1.0 rubrics.

Validators reject; rubrics rank. Use this when you want to know how good an output is on each axis, not just whether it passes a binary gate.

use prompt_eval_rubric::{Rubric, RubricSet};
use serde_json::json;

let length_ok = Rubric::new("length", |out, _ctx| {
    let n = out.len();
    if n >= 10 && n <= 200 { 1.0 } else { 0.0 }
});

let has_json = Rubric::new("has_json", |out, _ctx| {
    if serde_json::from_str::<serde_json::Value>(out).is_ok() { 1.0 } else { 0.0 }
});

let set = RubricSet::new(vec![(length_ok, 0.4), (has_json, 0.6)]).unwrap();
let report = set.evaluate(r#"{"key": "value"}"#, None);

assert!(report.overall > 0.5);
println!("overall: {}", report.overall);
for s in &report.scores {
    println!("{}: {:.2} {:?}", s.name, s.value, s.reason);
}

Structs

Report

Aggregate result from RubricSet::evaluate.

Rubric

One scoring axis, wrapping a callable (output, context) -> score.

RubricSet

Aggregate multiple rubrics with optional weights.

Score

One rubric’s score for one output.

Enums

RubricSetError