Validation report — OpenLiDARViewer v0.6.0-alpha.1

This report states, soberly, what v0.6.0-alpha.1 validates and what it does not. It is the human-readable companion to the machine-readable claim register (docs/validation/claim-register.yaml) and the alpha review response (docs/_audit/v0.6-alpha-blocker-response.md).

The v0.6 cycle's changes are in streaming decode, session import, measurement-unit honesty, and a shared project-frame foundation. The terrain and contour correctness claims are inherited unchanged from v0.5.9 — the alpha wave did not touch those algorithms — so their evidence remains as recorded in VALIDATION_REPORT_v0.5.9.md. This report covers the alpha-specific surface on top of that.

Evidence ceiling

Unchanged from v0.5.9: no product is validated above internal evidence. On the E0–E6 ladder nothing is at or above E4 (cross-implementation independence); synthetic known-truth checks reach E3. The alpha's new correctness guards are validated at E2–E3 (unit tests against constructed inputs), not against an independent reference implementation.

What was tested (alpha wave)

Run with npm run test:unit, test:export, test:terrain, test:ui, test:slow, plus npm run test:file <path> for a single file.

• Streaming non-finite refusal. tests/streamingFiniteGuard.test.ts verifies COPC/EPT decoders refuse a node whose transform is non-finite, whose finite-but-extreme transform overflows a coordinate to Infinity, or whose EPT-binary float source carries a NaN — with a structured malformed-file error, never NaN to the GPU.
• Session scan-identity guard. tests/sessionScanIdentity.test.ts verifies matchSessionToScan returns strong / partial / conflict on extents (primary), point count (corroborating, reduction-tolerant), and name/CRS (disclosure), including the rename and device-reduced edge cases.
• Stockpile density-unit honesty. tests/stockpileVolume.test.ts / tests/stockpilePresenter.test.ts verify HIGH confidence is withheld when the horizontal unit is unknown and the density row is labelled pts/unit² (unit unknown) rather than claiming m².
• Project-frame transform math. tests/projectSpatialFrame.test.ts verifies the source↔project↔world round-trip and the Float32 sub-mm residual bound.
• Progressive EPT attach. tests/eptStreaming.test.ts verifies first-paint-then-continue parity with the full walk and that a persistently-failing fetch terminates (no allocation loop).

Whole-suite evidence, run locally at the alpha head commit (not yet a Git tag): unit 2,652 passed / 16 skipped, export 550, terrain 1,181 / 18 skipped, ui 429, slow 505; build-contract 11; live/obfuscated build passed; production dependency audit 0 vulnerabilities. The full e2e suite passed locally here (161 passed / 5 fixture-skipped / 0 failed) — the gating browser evidence is a green GitHub Actions run on the tagged commit, which is pending (see "What was NOT tested").

What was NOT tested (and is staged, not claimed)

• The shared project frame is a foundation only. ProjectSpatialFrame is imported by tests and documentation, not by the running viewer. Multi-dataset overlay, Compare Studio, and cross-layer picking do not yet mount layers through it. See KNOWN_LIMITATIONS_v0.6.0-alpha.1.md.
• The anti-thrash streaming-selection option is opt-in and unwired. Its logic is unit-tested; its visual effect on flicker is unverified because it needs a browser and is not enabled in this build.
• Browser behaviour on GitHub CI is not part of this archive's evidence. The e2e suite passed locally; a green GitHub Actions run on the exact tagged commit is required before publication and is not asserted here.

Reproducing

npm ci
npm run test:release     # typecheck, lints, live build, all buckets, smoke
npm run test:e2e         # full Playwright suite

Verdict

The alpha's new correctness guards are validated at the internal-evidence ceiling; the inherited terrain/measurement claims stand as in v0.5.9. The project-frame runtime integration and the browser-verified items are explicitly out of scope for this archive's evidence and are documented as staged.