Skip to main content

qualia_core_db/inference/conditioning_eval/
report.rs

1//! Paired comparison reporting for baseline vs compiled conditioning.
2
3use super::scoring::TaskScore;
4
5#[derive(Debug, Clone)]
6pub struct PairedComparison {
7    pub task_id: String,
8    pub baseline_score: TaskScore,
9    pub compiled_score: TaskScore,
10}
11
12impl PairedComparison {
13    /// Paired delta between baseline and candidate.
14    pub fn delta(&self) -> f32 {
15        self.compiled_score.composite_score() - self.baseline_score.composite_score()
16    }
17}
18
19#[derive(Debug, Clone)]
20pub struct EvaluationReport {
21    pub run_id: String,
22    pub comparisons: Vec<PairedComparison>,
23}
24
25impl EvaluationReport {
26    /// Creates a default `EvaluationReport`.
27    pub fn new(run_id: impl Into<String>) -> Self {
28        Self {
29            run_id: run_id.into(),
30            comparisons: Vec::new(),
31        }
32    }
33
34    /// Mean improvement across comparisons.
35    pub fn average_improvement(&self) -> f32 {
36        if self.comparisons.is_empty() {
37            return 0.0;
38        }
39        let sum: f32 = self.comparisons.iter().map(|c| c.delta()).sum();
40        sum / self.comparisons.len() as f32
41    }
42}