perfgate-paired

Paired benchmarking statistics for A/B comparison.

Part of the perfgate workspace.

Overview

Provides statistical analysis for paired benchmark data where each measurement consists of a baseline and current observation from the same experimental unit. Uses paired t-test methodology with 95% confidence intervals.

Key API

compute_paired_stats(samples, budget) — compute summary stats from paired samples → PairedStats
compare_paired_stats(stats) — compare with confidence intervals → PairedComparison
summarize_paired_diffs(samples) — summarize the distribution of differences
PairedError — error type (re-exported from perfgate-error)

Statistical Methodology

Paired t-test with conservative t-values (2.0 for n < 30, 1.96 for n ≥ 30)
95% CI: mean ± t × (std_dev / √n)
Significance: CI does not span zero

Example

use perfgate_paired::{compute_paired_stats, compare_paired_stats};
use perfgate_types::{PairedSample, PairedSampleHalf};

// After collecting paired samples...
// let stats = compute_paired_stats(&samples, None)?;
// let comparison = compare_paired_stats(&stats);
// println!("Mean diff: {:.2}ms", comparison.mean_diff_ms);
// println!("Significant: {}", comparison.is_significant);

License