Coverage Gate
Each mandatory check carries its own capability, evaluation and evidence state. A “ready” verdict is structurally impossible while any required row is uncovered — the tool cannot silently skip a rule.
Audit engine live in production — not a mock-up
Every IFC version checked automatically — the way software teams run tests before a release. Applicant-side, IFC-native, and every verdict carries its own evidence.
Illustrative run against the Swedish rule pack in development. Today the engine runs in production on Russian design documentation.
01 — Problem
The check is manual, the codes move faster than the engineers, and the cost lands on the applicant.
02 — Multi-Agent Consensus
Watch 18 domain-specialized AI reviewers stream proof-carrying verdicts directly into text blocks on the submission document — forming structural consensus across all building codes.
All EI 60 rated compartment walls shall extend continuously to slab undersides. Maximum egress path length to safest stairwell: 27.4 meters.
Main corridor clear width is 1350mm (Min 1200mm). Ramp 3B gradient is 1:11 (8.9%) — landing length requires engineering verification.
Primary steel frame HEB 300 columns meet C3 live load capacity under EN 1991-1-1. Provenance graph anchored to ISO 19650 metadata.
Duct velocity 6.2 m/s maintained with 450mm structural clearance. Acoustic insulation 52dB (SS 25267). Energy Class A+ (EPBD).
03 — 3D Spatial Audit
Inspect building geometry in 3D. The automated scanner plane passes through structural, architectural, and MEP layers, continuously validating compliance rules against IFC entity attributes.
04 — IFC Federation
From raw architectural CAD and IFC geometry to correlated rule execution. Every finding connects back to its precise source in the model.
Overlays live IFC data structures, 3D structural analysis, and municipal building codes onto unified compliance maps.
Types joins across architecture, structural engineering, and MEP routing. Finds clashes and code violations before construction starts.
05 — Solution
Two mechanisms make the verdict trustworthy. Both are patent pending.
Each mandatory check carries its own capability, evaluation and evidence state. A “ready” verdict is structurally impossible while any required row is uncovered — the tool cannot silently skip a rule.
Cross-domain findings are typed joins over IFC entities with a provenance trail back to source facts — so a finding can be re-derived, not just trusted.
Codes move: national, state and municipal amendments land continuously. Any past review can be replayed against the rules as they stood on the submission date.
06 — Proof
07 — Market
Capital and regulation arrived in 2026; the installed base did not. That gap is the entry window.
08 — Competition
“Upload a PDF and let AI find code violations” is a crowded lane. Neutral IFC plus jurisdiction replay is not.
| Tool | IFC-native | Code corpus | Applicant-side | Cross-jurisdiction | Provable coverage |
|---|---|---|---|---|---|
| Solibri / BIMcollab BIM QA, rules you author yourself | ● | · | ● | · | · |
| UpCodes / Helonic code content + PDF reading | · | ● | ● | · | · |
| Kestrel strong, but Revit-only | · | ● | ● | · | · |
| CodeComply / CivCheck sells to city halls, not firms | · | ● | · | · | · |
| AiTech (SE) Sweden — municipality-facing | · | ● | · | · | · |
| BIM Grade the whole row | ● | ● | ● | ● | ● |
Fragmentation is the moat — there is no single global building code, and that is exactly why a jurisdiction-replay engine compounds.
The engine runs today on real, paid project documentation. The Swedish rule pack — BBR (Boverkets byggregler, the Swedish building regulations) and PBL (Plan- och bygglagen, the Planning and Building Act) — is the first Western target; DACH and US follow.