Not "trust us" — benchmarked against independent, already-verified results, with a record generated by the benchmark itself.
Assessments across the engine carry a verification record generated by the benchmark itself. When a module is removed from the engine, its verification badge is withdrawn automatically.
A bearing capacity without a confidence interval is an opinion with a unit attached, so ORB reports both the value and its 95% interval.
Most tools report only the 763.
A gap carries the same fields as any judgement: which domain, why, and what would be required to close it. ORB tallies them across every real file, so the accumulated backlog of genuine gaps decides what gets built next.
A Dutch groundwater table, quietly computed in Nairobi, is worth less than nothing.
Under every claim sits the kernel’s event log. Each create, update and delete writes an immutable event: who, what, when, the old value, the new value and the reason. And every property carries its own provenance — value, unit, source, reliability and validity. That is how a figure years from now still points back to where it came from.
Events are read-only to everyone; only the engine writes them, and approved work is locked. So the record that justified a decision is the record it is later checked against.
ORB's judgements rest on public, authoritative data. Some layers are global — the same for every place on earth: the satellite imagery and terrain you see, the change record since 2015, worldwide building footprints and the protected-area register. Others are national, and plug in per country by configuration — the geotechnical surveys, cadastres and registers a country built for itself. Which source applies, and which standard governs, is a setting per country and per domain: a new country is a configuration, not a rebuild. Where a source is absent, the assessment says so and drops to a desk screening rather than guessing.
| Domain | Source & coverage | Method / standard |
|---|---|---|
| Terrain & imagery | Global — Esri World Imagery with WorldElevation3D. Every point on earth at one resolution — the globe you are looking at. | The base layer, everywhere |
| Change over time | Global — Sentinel-2 L2A · ESA Copernicus via Element84 STAC · 10 m resolution · a fresh pass roughly every 5 days · a continuous record since 2015. | Pettitt (1979) changepoint over the whole series |
| Ecology | Global — WDPA protected areas (UNEP-WCMC / IUCN), covering every country on earth. | Protected areas · distance and overlap |
| Buildings | Continental — Google Open Buildings across Africa, South Asia and Latin America; national cadastres (footprint and year built) where a country maintains one. | Proximity & density · receptor for noise and vibration |
| Population | Continental — GRID3 and CIESIN settlement layers, complemented by national statistics. | Population and settlement extents |
| Subsurface | Country — 3D voxel models and boreholes where a national survey exists (down to 50 m); regional soil datasets elsewhere. | Voxel column per point, top 50 m below surface |
| Soil mechanics | Country — CPT soundings and voxel-derived profiles where available. | Koppejan · NEN 9997-1 §7.6 · Terzaghi-Buisman |
| Hydrology | Country — groundwater and precipitation records where a national source is connected. | Groundwater within 500 m — absence is reported |
| Geochemistry | Country — contaminated-site registers where a country keeps one. | Contaminated-site register · NEN 5740 |
| Vibration | Country — computed on the building set, with damping read from the subsurface model. | SBR guideline A · receptor from the building set |
Every source here is public: TNO, PDOK, BRO, RWS, ESA Copernicus. A phase-1 result is a desk screening, and where a phase-2 anchor to local measurement is absent the assessment falls back to phase 1 and marks itself accordingly, so the resolution drops visibly rather than quietly. None of this replaces a CPT under NEN-EN-ISO 22476-1 or a NEN 9997-conform design by a geotechnical adviser — a limit the report itself records.
