Clash detection
A 5 mm tolerance can hide the clashes that matter
On a renovation in Dubai, the contractor's clash tests ran correctly. I reproduced every one of their 305 open results. The problem was what the tests were set up to look for.
Their structure against mechanical test used a 5 mm tolerance. That sounds harmless, but on that one model pair it hid 28 real clashes. Eight results the contractor had marked resolved were still clashing when the test ran again at 0 mm, and 147 results overlapped by more than 25 mm. A 50 mm clearance test, which nobody had run, returned 941 items.
None of this is a software problem. Navisworks did exactly what it was told. The gap was in the clash matrix: which pairs are tested, at what tolerance, with what clearance, and who owns each result. Without an agreed matrix, a clash report measures the settings rather than the model.
My rule now is simple. Before reviewing anyone's clash results, rerun one model pair yourself at 0 mm and again with a realistic clearance. If the numbers match theirs, you can trust the rest of the report. If they do not, you have found the first finding.
Model review
177,781 clashes is not a finding
On a hospital review, the raw clash count was 177,781. A number like that is useless to a design team. Nobody can act on it, and it hides the handful of problems that would stop a pour.
The work is in the reduction. Results are grouped by element and location, tested against 49 rules that reflect how each discipline is built and maintained, and reduced to findings a person can close: 1,835 in this case, 317 of them critical or major.
Then every finding is classified by who resolves it. A design stage issue goes back to the designer, an interface issue needs two trades in the same room, and a construction stage item belongs to the contractor. About 36 percent sat at design stage, which is exactly where they are cheapest to fix.
One example made the point better than any chart: an 1,100 by 950 mm duct running straight through a 1,000 by 1,000 mm concrete column at basement level. One finding, one owner, one correction before anything is built.
Digital transformation
The 140 checklists were one form
A compliance team asked for a dashboard across 21 countries. The brief described 140 Excel checklists. Before designing anything, I read the files.
Three things changed the whole project. The 140 checklists shared one structure, and about 62 percent of their content was identical international standards, so the build became one form engine plus a country overlay. The same workbook defined compliance status three different ways, so any chart grouped by severity would have counted categories that did not exist. And the KPIs the client wanted needed data the checklists never captured: a finding had no concept of being closed.
None of that was visible in the brief. It was only visible in the files. Fixing it first is what made the dashboard mean anything.
Digital transformation usually gets described as choosing the right platform. In my experience it is mostly reading what people actually use, finding where it contradicts itself, and designing the data before the screens.