Assurance and evidence
What a skeptical reader can check themselves.
| Claim | Evidence | How to re-check |
|---|---|---|
| Tests pass | test-results/test-run.tap (153 pass, 0 fail) | npm test |
| Lineage matches labels | test-results/gold-results.json (54/54) | npm run proof |
| Read-only | audit: tree hash identical before/after | lineagesutra audit --out x |
| Offline | audit: core/reporting import no network or subprocess modules | open packages/core |
| Deterministic | audit: same seal, any file order | audit |
| Tamper-evident | audit: three mutation kinds fail verify | lineagesutra verify |
| Robust | audit: 3,000 seeded fuzz inputs, 500-deep nesting, 5,000 columns | audit |
| Fast | audit: undefined ms / undefined ms | audit |
| Package integrity | audit: 25 skills, 20 roles, schemas, no tokens | audit |
Sample artifacts
samples/audit-report.html,.md,.jsonsamples/cli-*.txtreal command transcriptstest-results/environment.json
What is not evidenced
Independent review, third-party MCP client runs, native IDE activation, real-world dialect accuracy, signed provenance. These are listed as NOT_RUN in every audit.
Verdict scale
PASS = all executed checks passed. Not a certification.