Skip to main content
In progressCODECHECKRevitAccessibilityCode ReviewIFC

A Seoul-based architecture & engineering firm

CODECHECK: Automated Accessibility Code Review inside Revit

Geometry-based accessibility (barrier-free) code review running directly inside Revit. CODECHECK triages every sub-criterion for automatic judgability and measures live model geometry against Korea's Barrier-Free code, backing each verdict with the extracted value, formula, threshold, and clause rather than trusting model parameters alone. Joint PoC underway under MOU.

Client
A Seoul-based architecture & engineering firm
Location
Korea
Environment
Revit 2024 · IFC
Status
In progress

Context

Our client is a Korean architecture and engineering firm delivering large residential projects in Revit. In Korea, accessibility compliance (the Act on Guarantee of Convenience Promotion for Persons with Disabilities) is reviewed against a checklist of 152 sub-criteria — and in most firms, that review is done by hand, by the few people who actually know the regulation.

The bottleneck

Manual accessibility review has three structural problems. It doesn't scale: every design change restarts the review. It arrives too early or too late: structural-stage models carry no finishes, so a naive pass/fail verdict flips once finishes are applied. And it can't be trusted blindly: parameter values in real models drift from the actual geometry, so a review based on properties alone measures the wrong thing.

The goal

Verdicts a practitioner can trust without leaving Revit. That meant three requirements: measure geometry, not parameters; make every verdict show its reasoning — extracted value, formula, threshold, clause; and work on the model as it is, telling the user what to add when an item can't be judged yet, instead of failing silently.

Workflow & standards fit

The rule base covers the Korean Barrier-Free enforcement rules (Annex Table 1) and is designed to layer the firm's internal design guide on top — the same engine, two rule sources. Because structural-stage models understate finished dimensions, verdicts are three-band rather than binary: fail / margin-critical / clear, using the firm's finish-thickness allowances. The checklist lives in an external knowledge base, so a regulation change is a data update, not a code change.

What we built

We first triaged all 152 sub-criteria for automatic judgability against the live model — possible now, conditional, blocked (with the exact modeling step that unblocks it), or out of scope for geometry-based checking. Then we measured a live 4-tower, 29-storey residential project: 26 ramps across 29 segments, 23 stairs, 42 doors, 19 walkway subregions and 3,007 structural beams — every verdict backed by measured geometry, with parameter-vs-geometry mismatches flagged. The delivery vehicle is a Revit add-in (in development): scan the model, see which items are judgable right now, run the check, and click any finding to zoom to it in 3D with a full evidence card. Review views persist after the add-in closes; every run is snapshotted so the next one reports what got fixed and what newly broke.

Image pendingMockup screenshot pending — v3 mock, triage view.
Model scan & triage · possible / conditional / blocked, per sub-criterion.
Image pendingMockup screenshot pending — v3 mock, evidence card view.
Finding detail · zoom to element in 3D, with extracted value, formula, threshold, clause.

Engagement

The joint PoC runs on a weekly demo cadence — screens over documents. Within the first week after data handover we returned the full 152-item triage plus measured verdicts across five categories. The practitioner feedback loop is already working: a reviewer caught that stair clear width must be measured between railings, not family bounds — the rule was corrected the same week.

Image pendingMockup screenshot pending — v3 mock, project review view.
Project-level review · verdicts across ramps, stairs, doors, walkways, and beams.

Outcome

As of August 2026, all 152 sub-criteria have been triaged, with first-wave automatic coverage established. Five categories have been measured and judged on a live project model, and the judgment logic has been validated against an independent IFC pipeline. Add-in implementation is underway; a client quote is pending publication approval.

Image pendingMockup screenshot pending — v3 mock, run comparison view.
Run comparison · each run snapshotted, so the next one shows what got fixed and what newly broke.

Tech stack

RevitC#Revit APIWPFPythonIfcOpenShellJSON rule base

Have a similar bottleneck?

We start with a field interview, define the pain point precisely, and get to a working build fast.

Talk to sales