Test report
Recorded 2026-11-16T18:03:17.735Z on win32/x64, Node v24.15.0. Raw output: test-results/test-run.tap.
| Measure | Value |
|---|---|
| Tests | 157 |
| Passed | 153 |
| Failed | 0 |
| Skipped | 4 (symlink tests that Windows forbids without privileges) |
Test areas
- Tokenizer, parser, scan, determinism, file-order independence
- Gold corpus: 54 cases, 148 expected edges, precision 100%, recall 100% (author-labelled)
- Impact, upstream, compare (DriftRadar), contract, dbt
- Property test: 200 random graphs, adding edges never shrinks confirmed impact
- Fuzz: 1,500 seeded token streams in tests; 3,000 more in the audit
- CLI: formats, exit codes, gates, audit outputs, npm-bin symlink behaviour
- MCP: unit and real stdio round trip
- Security: tamper detection, symlinks, caps, XSS, injection text
Gold corpus detail
54 cases tagged basic, join, expr, indirect, cte, subquery, setop, wildcard, ddl, naming, postgres, tsql, mysql, dbt, adversarial, negative. Full per-case results: test-results/gold-results.json.
Audit
28 passed, 0 failed, 7 not run. Open samples/audit-report.html.
Performance (this machine, synthetic)
2,000 chained views: undefined ms median. 20,000-dataset impact: undefined ms median.
Not covered
Real-world dialect corpora, third-party MCP clients, native IDE skill activation, independent review. CI results for Windows, macOS and Linux are on the GitHub Actions page.
Reproduce
npm test
npm run proof