Available on crate feature
eval only.Expand description
Human annotation workflow via JSONL export/import.
Provides JSONL-based file format for exporting evaluation cases for human review and importing human verdicts back into the system.
§Format
Each line in the JSONL file is a JSON object representing an AnnotationRecord.
On export, the verdict field is null; on import, annotators fill it in.
§Example
ⓘ
use adk_eval::annotation::{AnnotationStore, AnnotationRecord};
use adk_eval::schema::EvalCase;
use adk_eval::report::EvaluationResult;
use std::collections::HashSet;
// Export cases for human review
AnnotationStore::export(&cases, &results, "annotations.jsonl")?;
// Import annotated verdicts
let valid_ids: HashSet<String> = cases.iter().map(|c| c.eval_id.clone()).collect();
let (records, warnings) = AnnotationStore::import("annotations.jsonl", &valid_ids)?;Structs§
- Annotation
Record - A single annotation record for human review.
- Annotation
Store - Manages JSONL export and import for human annotation.
- Human
Verdict - Human-provided verdict for an evaluation case.