use crate::output::OutputFormat;
use aether_evals::{EvalFilesReport, EvalOutcome};
use std::fmt::Write as _;
pub fn render(report: &EvalFilesReport, format: OutputFormat) -> String {
match format {
OutputFormat::Json => serde_json::to_string(report).expect("EvalFilesReport serializes"),
OutputFormat::Pretty => serde_json::to_string_pretty(report).expect("EvalFilesReport serializes"),
OutputFormat::Text => {
let mut output = String::new();
for eval in &report.evals {
if eval.passed {
let _ = writeln!(output, "✓ {}{}", eval.name, judge_score(eval));
continue;
}
let _ = writeln!(output, "✗ {}{}", eval.name, judge_score(eval));
for failure in &eval.failures {
let _ = writeln!(output, " - {}", failure.replace('\n', "\n "));
}
for line in eval.failure_context.iter().flat_map(|context| context.lines()) {
let _ = writeln!(output, " {line}");
}
}
let _ = write!(output, "\n{} passed, {} failed", report.passed_count(), report.failed_count());
output
}
}
}
fn judge_score(eval: &EvalOutcome) -> String {
eval.judge.as_ref().map(|judge| format!(" (judge score {:.0}%)", judge.score * 100.0)).unwrap_or_default()
}
#[cfg(test)]
mod tests {
use super::*;
use aether_evals::{EvalOutcome, JudgeCriterionSummary, JudgeSummary};
fn report() -> EvalFilesReport {
EvalFilesReport {
evals: vec![
EvalOutcome {
name: "passing".to_string(),
passed: true,
failures: vec![],
tool_calls: vec![],
judge: Some(JudgeSummary {
passed: true,
score: 1.0,
reason: "weighted score 1.00; all blockers met".to_string(),
criteria: vec![JudgeCriterionSummary {
id: "behavior".to_string(),
description: "correct behavior".to_string(),
blocking: true,
weight: 1.0,
threshold: 1.0,
score: 1.0,
passed: true,
reason: "correct".to_string(),
}],
}),
failure_context: None,
retained_workspace: None,
},
EvalOutcome {
name: "failing".to_string(),
passed: false,
failures: vec!["expected tool `bash` to be called".to_string()],
tool_calls: vec![],
judge: Some(JudgeSummary {
passed: false,
score: 0.0,
reason: "weighted score 0.50; one or more blockers failed".to_string(),
criteria: vec![JudgeCriterionSummary {
id: "behavior".to_string(),
description: "correct behavior".to_string(),
blocking: true,
weight: 1.0,
threshold: 1.0,
score: 0.5,
passed: false,
reason: "wrong".to_string(),
}],
}),
failure_context: Some("Task failure context\nWorkspace: /tmp/x".to_string()),
retained_workspace: None,
},
],
}
}
#[test]
fn text_render_marks_pass_fail_and_totals() {
let text = render(&report(), OutputFormat::Text);
assert!(text.contains("✓ passing (judge score 100%)"));
assert!(text.contains("✗ failing (judge score 0%)"));
assert!(text.contains("expected tool `bash` to be called"));
assert!(text.contains("1 passed, 1 failed"));
}
#[test]
fn json_render_is_machine_readable() {
let json = render(&report(), OutputFormat::Json);
let parsed: serde_json::Value = serde_json::from_str(&json).unwrap();
assert_eq!(parsed["evals"][0]["name"], "passing");
assert_eq!(parsed["evals"][1]["passed"], false);
assert_eq!(parsed["evals"][1]["failures"][0], "expected tool `bash` to be called");
assert_eq!(parsed["evals"][0]["judge"]["score"], 1.0);
assert_eq!(parsed["evals"][1]["judge"]["criteria"][0]["id"], "behavior");
assert_eq!(parsed["evals"][1]["judge"]["criteria"][0]["passed"], false);
}
}