Work record / Stories

Trusting a model because it looked finished

Before the blind second pass was a rule, I accepted a design set that read complete and was built on another address's plans.

I trusted a model because it looked finished, and that was before the blind second pass was a rule. On the Midtown tower our agentic system, which ran the architecture, turned out a design development set that came across as complete and internally consistent. I filed it as finished.

It was not done. We later went back to re-anchor the geometry to the real site plans, and the footprint was wrong. Geometry in the model had come from the plan set of a different address. Nobody had checked it independently against the source, so nobody caught it.

The cost was real: time lost redoing work I had already filed as finished in my head. It could have gone much further had it reached the city first. We rebuilt against verified plans and quit treating internal consistency as proof that a model matches the world.

It is part of why we now re-measure every takeoff and every model blind, by an independent pass, before calling any of it finished. Mine included.

Story details

All stories · Work record