The crowd finds it
The Millennium Bridge opened on 10 June 2000. That day 90,000 people crossed it, up to 2,000 standing on it at once, and it swayed 70 millimetres side to side.
They closed it on 12 June. Two days.
The engineering was sound. What nobody modelled was that people adjust their footing when the ground moves, and 2,000 people adjusting at once push in time with the sway.
The crowd caused the wobble. The crowd was the only thing that could have revealed it.
You could have walked ten engineers across that bridge every day for a year and found nothing.
This is the failure a pilot cannot find. Five people using the new process is not a small version of fifty people using it. Some problems are made of the scale itself, and they arrive on the day you stop testing and start operating.
It stayed shut for twenty months. The fix was 37 dampers, then 52 more, and £5 million.
Ask what your process does at full load. Not whether it works.