Detect flaky LLM eval cases across repeated runs. Per-case classification (AlwaysPass/AlwaysFail/Flaky) plus a flip-rate score.
use eval_flake_rs::detect; let runs = vec![ vec![("c", true)], vec![("c", false)], vec![("c", true)], ]; let r = detect(&runs);
Zero deps. MIT or Apache-2.0.