A lab prototype that had to become a manufacturable instrument.
The science worked. The instrument around it did not yet exist as a product — it could not be built twice the same way, serviced in the field, or documented well enough to be used clinically.
A product that exists once is not a product.
The team had spent two years proving the measurement worked. What they had was one instrument, assembled by the people who designed it, held together by knowledge that had never been written down.
They needed units that behaved identically, could be serviced by someone else, and came with the evidence a clinical setting would demand.
A benchtop diagnostic instrument used at the point of care. Samples are prepared and measured on the device, and the result has to be produced in minutes, by an operator who is not an engineer.
That places the demand on the whole system at once: fluid handling, sensing, timing, the operator interface and the record the instrument leaves behind.
Every part of the prototype assumed an expert was standing next to it.
Timing was tuned by hand for one set of hardware. Failures were recoverable only by someone who knew the internals. Nothing recorded what the instrument had actually done during a run.
The engineering problem was not the measurement. It was everything the measurement depended on, and none of it could be fixed in isolation.
We kept the proven measurement path and rebuilt the system around it, in an order that let the team keep testing throughout.
An instrument that can be built, serviced and defended.
Units now behave the same way as each other. Faults are visible and recoverable without the original engineers. The documentation exists because the architecture produced it, not because someone wrote it afterwards.