Skip to main content

qualia_core_db/inference/conditioning_eval/
scoring.rs

1//! Task output scoring and validation against ground truth.
2
3#[derive(Debug, Clone, Copy, PartialEq, serde::Serialize, serde::Deserialize)]
4pub struct TaskScore {
5    pub accuracy: f32,
6    pub citation_fidelity: f32,
7    pub constraint_satisfaction: f32,
8}
9
10impl TaskScore {
11    /// Weighted composite score.
12    pub fn composite_score(&self) -> f32 {
13        0.5 * self.accuracy + 0.3 * self.citation_fidelity + 0.2 * self.constraint_satisfaction
14    }
15}
16
17/// Score an output string against the expected outcome and evidence citations.
18pub fn evaluate_task_output(
19    actual: &str,
20    expected: &str,
21    citations_present: bool,
22    constraints_satisfied: bool,
23) -> TaskScore {
24    let accuracy = if actual.trim() == expected.trim() {
25        1.0
26    } else if actual.contains(expected) {
27        0.7
28    } else {
29        0.0
30    };
31
32    let citation_fidelity = if citations_present { 1.0 } else { 0.0 };
33    let constraint_satisfaction = if constraints_satisfied { 1.0 } else { 0.0 };
34
35    TaskScore {
36        accuracy,
37        citation_fidelity,
38        constraint_satisfaction,
39    }
40}