The most dangerous moment in a hardware programme is the one immediately after a prototype works. It feels like the hard part is over. In terms of engineering effort, roughly half of it is still ahead.
One unit hides variation
A prototype is usually built by the person who designed it, adjusted as it goes, with parts selected from whatever fitted best. Every one of those acts absorbs variation that production will not absorb. At volume, parts arrive at the limits of their tolerance, assembled by someone who has not seen the CAD, on a schedule.
Process introduces its own behaviour
Moulded parts shrink, and not uniformly. Painted surfaces build thickness. Adhesives cure at a rate that depends on temperature and humidity. None of these appear in a single hand-built unit, and all of them appear in a production run.
Closing the gap
- Build a pilot run using the real sequence and real fixtures
- Assemble deliberately from parts at tolerance extremes
- Have someone unfamiliar with the design build it from the instructions
- Record process parameters so they can be held, not rediscovered
- Define and measure yield before it matters commercially
The goal is a process where quality comes from the system rather than from supervision — because supervision does not scale, and the second production run rarely gets as much of it as the first.