@evidtrail/metrics

Deterministic metric calculation for versioned evidtrail repository snapshots.

What it calculates

• Repository change signals — fixed-horizon rapid retouch at 7, 30, and 90 days. Files without a complete observation window stay visible as tooRecent; rapid retouch is a file-level churn signal, not proof of a defect, wasted work, or causality.
• Trend — repository change signals over monthly or quarterly periods. Every period uses the same observation window; immature periods are reported but excluded from comparisons.
• Evidence coverage and autonomy cohorts — declared, inferred, and missing provenance, plus optional comparisons by autonomy mode. Cohort comparisons are withheld when the evidence cannot support an honest comparison; defaultMode is a prior for cohort membership, never evidence.
• Line survival — optional git blame input reports attribution for lines alive in the sampled tree. The surviving-line share is exact for that input; the derived survival rate is explicitly approximate because deleted lines are unavailable to blame.
• Outcome correlation — git-visible reverts and hotfix-pattern commits, expressed against each cohort's authored-commit base rate. The results are descriptive, not causal, and do not include incidents or external security findings.
• PR acceptance — optional forge PR input reports merged versus closed-unmerged PRs, including attribution and autonomy breakdowns. It is absent when no PR stream is supplied, never silently zero.

For why evidtrail does not calculate Merge Ratio from Git history, see Why there is no Merge Ratio in the repository README.

Usage

import { calculateMetrics } from '@evidtrail/metrics';

const metrics = calculateMetrics(commitStream);
const retouch30 = metrics.repo.persistence.rapidRetouch.find((r) => r.windowDays === 30);
console.log(`30-day rapid retouch: ${retouch30?.rate ?? 'unavailable'}`);
console.log(`Evidence coverage: ${metrics.attribution.coverage}`);

Pass blameStream from evidtrail blame or prStream from evidtrail fetch-prs to calculate the corresponding optional metrics. See the root README for the metric contracts, caveats, and CLI workflow.