The tolerance chain that couldn't be built

Here's a question that decides more money than almost anyone realises: when your work has to land on somebody else's work, do the two specifications actually agree on what close enough means? Because on the job I'm going to describe loosely, they didn't, nobody noticed until the steel arrived and the argument that followed ran well into six figures.

The setup was ordinary. A logistics unit, portal frame on pad foundations, one contractor casting the concrete and setting the holding-down bolts, another fabricating and erecting the steel. Two packages, two specifications, each written by a different designer and each perfectly reasonable on its own terms. The concrete spec gave the foundation contractor a tolerance on bolt positions of the sort concrete work has always been given, because concrete is poured into the ground by human beings in weather and you cannot hold it to watchmaker's limits. The steel spec, meanwhile, assumed baseplates would land on bolts positioned to the much tighter limits that fabricated steel is drawn to, because steel comes off a machine in a shed and can be.

Both documents were right. Both contractors built to their own spec and could prove it. And the two tolerances, stacked together at the interface where one trade's work met the other's, could not both be true. Bolts cast comfortably inside the concrete tolerance still sat outside what the baseplates could accept. The frame turned up, the first plates wouldn't land and everything stopped while the surveyors argued about whose numbers were whose. The remedial answer was the usual ugly menu, engineered justifications leading to expensive remedial works, all of it costing money that had never been in anybody's price and all of it happening at the worst possible time, with a crane on hire and an erection gang stood watching.

Then came the second argument, the expensive one: whose job had it been to spot the clash? The foundation contractor said, reasonably, that they'd built what their documents said. The steel contractor said the same. The main contractor, who had procured both packages and was the only party holding both specifications, felt strongly that somebody else should have read them side by side. And here's the uncomfortable truth of it: contractually the answer varies job by job, but practically the party who suffers is almost always the one with the least paper. The stopped gang, the standing crane, the delay to everything downstream, those costs land first on whoever can't show they raised the question in time.

Because that's the real lesson. This class of problem, two specifications that are individually fine and jointly impossible, is invisible inside either document. It only exists at the interface and interfaces are precisely where nobody's reading concentrates, because each designer owns their own document and, sometimes, no one is paid to own the seam between them. The fix costs almost nothing at the right moment: read the two tolerance regimes side by side before anything is cast and, where they can't both be true, raise it in writing as a question for the designers, early, when it's a drawing change rather than a drilling exercise. On an NEC job that's an early warning and it's precisely what early warnings are for. On any job it's a dated record that you saw the seam and asked, which is the difference between being the party who flagged the problem and the party who inherited it.

Every job you're on right now has these seams. Where your work lands on someone else's, has anyone read both documents at once? If the answer is nobody, then the question is only whether the seam happens to be buildable, and that's not a plan, that's luck.

Details changed to protect the parties; the pattern is real.