Everything here comes out of building and breaking a real CFD pipeline. No thought leadership, no predictions — measured numbers from runs that went wrong, and what it took to make the wrongness visible.
A completed simulation, a successful solver exit, sixteen green checks, and an issued report — for a compartment that reached 0.6 bar. How a smoke model seals itself, why nothing catches it, and the three separate defects we found while fixing the first one.
Read the case study →Our compute cost model under-predicted a real run by 2.8×. Not hardware, not mesh count — every calibration point predated the physics we now ship, so the table described a solver doing strictly less work than the one we run. What we measured, what we threw out, and the mistake inside the fix.
Read the method →Six checks had been quietly not running for months. The system reported it on every affected run, in plain English, and we read it every time — as a warning among warnings. Reporting a problem and communicating one are not the same act.
Read the case study →We declared envelope leakage, the deck contained it, and the pressure trace came back identical to the run with none at all — to five significant figures. What the solver actually requires, why it stays silent, and the two ways a leakage figure goes wrong even when the plumbing is right.
Read the method →Every storey above the fire read exactly ambient temperature and full visibility. Not because the smoke control worked — because a horizontal mesh boundary had severed the shaft at slab level. Why horizontal cuts seal and vertical ones do not, and what the signature looks like in someone else’s report.
Read the method →Approved Document B has no force in Scotland. Our reference library recorded that correctly and our check register cited it anyway, on every building. The correct fact sat in the codebase the whole time with nothing joining it to the decision that depended on it.
Read the piece →Why write about our own failures? Because in a liability profession the useful question is never whether a tool works when it works — it is what happens when it does not, and whether anyone finds out. Every piece here exists because something went wrong in a way that was invisible, and the fix is now a check that blocks its own author's work. If that is the sort of thing you would rather know before buying, you are the reader we are writing for.