OpenLiDARViewer v0.5.3

A hardening patch on the v0.5 line: two survey epochs can now be aligned before they are compared, the viewer installs and runs offline, and the evaluation behind the uncertainty claims reproduces with one command — on top of seventeen correctness fixes from two audit passes. Still browser-native and local-first. Nothing is uploaded.

Align two epochs before comparing them

The planar alignment core that shipped in v0.5.2 is now wired into change detection. Before two epochs are compared, the after cloud is coarse-registered onto the before cloud — yaw and horizontal shift only, because a real vertical change is the signal and must be preserved — and the fit is reported: the shift, the yaw, and the RMS residual appear in the compare result. A fit whose residual exceeds the gate is refused and the clouds are compared as-is, so alignment never invents a shift it cannot stand behind. The alignment reports in true metres on any projected CRS, applies in double precision so a georeferenced survey keeps its accuracy, and refuses geographic (degree) frames outright rather than fitting in a space where the numbers mean nothing.

Install it, run it offline

The viewer is now an installable PWA. A service worker caches the app shell, so after the first visit it opens and runs with no network, and it can be installed as a standalone app. The worker is local-first by construction: it caches only the same-origin app shell and never touches a dataset request, so opening a remote scan still goes straight to its source and nothing you load is stored.

A compass you can keep

The on-canvas compass from v0.5.2, previously reachable only through the ?viewcube=1 URL flag, can now be toggled from the command palette ("Toggle compass"), and the choice persists. It stays off by default so it never overlaps the panel columns; ?viewcube=1 and ?viewcube=0 still force it.

Reproduce the evaluation

npm run repro runs the real analysis cores over deterministic synthetic fixtures with analytic ground truth and writes a metrics table plus reliability figures: the epoch-registration vertical-bias result (horizontal-only registration preserves a uniform vertical change while a full-3D fit absorbs it), planar-alignment recovery, stockpile ±-band coverage against the nominal 0.68, and report-digest determinism. The coverage and bias checks also run as unit tests in CI, so the uncertainty claims are tested rather than asserted. REVIEWER_QUICKSTART.md gives the clone → test → repro → verify path.

Fixes

Seventeen correctness fixes land in this release — nine terrain/profile hardenings and eight fixes rated critical in a baseline audit of the v0.5 line. The theme is the project's usual one: a number you are shown is either right or refused with a reason.

Terrain and profiles:

• Geographic (lat/lon) scans no longer collapse to a one-cell analysis grid — the grid-cell floor is now unit-aware instead of treating 0.25° like 0.25 m.
• Slope, aspect, and hillshade on geographic grids use the true world latitude for the cos φ correction everywhere (it was silently degrading to the equator), on both the CPU and GPU paths, with the GPU equivalence gate extended so a kernel that ignores the correction can never pass.
• One percentile convention. Every reported p95 now uses the same type-7 quantile (the NumPy/R/Excel convention), so it is reproducible against standard tools.
• Contours are correct on fine and degree-denominated grids — the stitching tolerance scales with the grid cell, the coarse-interval check uses an exact level-crossing test, and saddle cells use the standard cell-average rule.
• The ground-filter despike now fires on small cells instead of silently keeping the lowest (blunder-included) return.
• Vertical grade is signed — a straight-down pair no longer reports an infinite climb.
• The terrain worker clamps an oversized point count instead of throwing and silently falling back to the slower path.
• Measurements on a geographic (degree) CRS are refused with a clear reason, not mislabelled in metres — pure-vertical heights and unit-free angles keep their ordinary grade.

Change detection, editing, sessions, and offline:

• Applying epoch alignment no longer injects up to half a metre of round-off jitter into a georeferenced survey; the transform is applied in double precision and the cloud keeps its origin.
• Alignment refuses geographic (degree) frames — a fit in lon/lat space could worsen registration while reporting success; reproject first.
• Alignment reports metres on any projected CRS — a 10 ft shift no longer reads "10.00 m".
• Geographic epochs refuse cut/fill volumes instead of printing degree areas as m³; the Δz statistics remain valid and are still reported.
• Lasso reclassify can no longer edit points you cannot see — points hidden by the clip box or a hidden class are excluded, matching the click-pick behaviour.
• After a classification edit, the on-screen analysis says it is stale ("results reflect the previous classification — re-run Analyse") instead of presenting the old result as current.
• The clip box survives a session round-trip — it is exported, restored without being clobbered by the reveal-time auto-frame, and reflected in the Clip panel.
• Visiting another page can no longer poison the offline app shell — the service worker refreshes the cached shell only from a clean scope-root navigation, never from a 404, a redirect tail, or a secondary page.

Known limitations

• The confidence figures and quality grades describe the delivered data; they are not a survey-grade certification. Treat terrain products, exports, and epoch comparisons as deliverable-ready only when the assessment reads Good, and validate against ground control where survey-grade accuracy is required.
• On a geographic (degree) CRS, lengths, areas, grades, profiles, volumes, and epoch alignment are refused with a stated reason rather than guessed — reproject to a projected CRS to measure.
• Epoch alignment is a coarse registration (yaw + horizontal shift) by design; it is not a substitute for survey co-registration.
• Offline installation requires a production secure origin; the first visit needs the network.

Verify this release

npm ci
npm test          # the full unit + integration suite
npm run repro     # writes benchmarks/out/metrics.md and the reliability figures

The release gate for this build ran the type check, the lint guards, both builds, the bundle budget, and all five test buckets green — 4,022 tests passed (30 skipped) across 352 test files. The figure step of npm run repro needs Python + matplotlib; the metrics table is written without it.

See CHANGELOG.md for the full list.

Deploy

Static files. Host anywhere (GitHub Pages, Netlify, any CDN). The deploy zip extracts with web-safe permissions (644 files, 755 directories) and carries index.html plus assets/ at the archive root, along with .htaccess and _headers for host-side security headers.