Reproducibility

This page is the single entry point for reproducing OpenLiDARViewer's build, tests, and reported analysis figures. It stitches together the pieces that also live in README.md, REVIEWER_QUICKSTART.md, and docs/benchmarks.md.

Environment

• Node.js 22 (pinned in .nvmrc and package.json engines; CI runs Node 22).
• Install exactly the locked dependency set with npm ci (uses package-lock.json; do not use npm install for a reproducible run).

nvm use            # picks up .nvmrc (Node 22)
npm ci

Commands

Purpose	Command
Type check	npm run typecheck
Unit + integration tests (partitioned)	npm run test:unit && npm run test:export && npm run test:terrain && npm run test:ui && npm run test:slow
Plain build + chunk-isolation contract	npm run test:build
Production (obfuscated) build	npm run build:live
Bundle budget check	npm run check:bundle
Full release gate (all of the above + lints + plain & live smoke)	npm run test:release
Regenerate analysis/benchmark figures	npm run repro
Package source + deploy archives	npm run package

The authoritative deterministic gate is npm run test:release (typecheck, lints, all test buckets, both builds, bundle budget, plus plain and live-obfuscated smoke). The broader browser matrix that a device varies — the full end-to-end suite and the mobile responsive job — runs in CI (.github/workflows/ci.yml), not in test:release; RELEASE_CHECKLIST.md lists that browser battery for a tagged release.

Determinism

• The analysis reproduction pack (tests/reproPack.test.ts, run via npm run repro) is fully deterministic: every input is generated from a fixed seed (an LCG plus Box–Muller), so the emitted metrics under benchmarks/out/ reproduce bit-for-bit on any machine. The report digest is content-addressed (tamper-evident).
• Test fixtures under tests/ are synthetic and seed-generated (see scripts/make-*.py and tests/fixtures/FIXTURES.md), or explicitly licensed (see THIRD_PARTY_NOTICES.md).
• Build identity is reproducible: the build honours SOURCE_DATE_EPOCH, and when git metadata is unavailable the commit is reported as unknown rather than fabricated.

Not deterministic (by nature)

• GPU/browser performance figures in docs/benchmarks.md depend on hardware, driver, and browser, and are described there as field observations, not a formal benchmark. They are not part of the reproducible metric set above.

Expected outputs

The committed reference outputs live under benchmarks/out/ (metrics.{json,md}, calibration, and registration-bias files). npm run repro regenerates them; a clean run reproduces the committed values.