The Batch That Was Rejected

The Challenge

The batch passed every check that existed at the time.

Dimensions were within tolerance. Finish was consistent. The material was correct and correctly sourced. Any single piece, handed to an inspector, would have been approved without hesitation.

Laid out together, they were obviously wrong. The colour drifted across the run — not enough to fail a measurement, easily enough to see when two pieces from opposite ends of the batch sat side by side.

The Process

The drift came from a substitution made three steps upstream and disclosed correctly. A component supplier had changed a finishing compound and told us, and the change had been assessed against the specification, and it passed.

What it did not pass was a test that did not exist: consistency across a run, rather than conformance of a unit.

Reconstructing how it happened took eight days and involved nobody hiding anything. That is the part worth recording. Every individual decision was defensible and documented, and the outcome was still forty unusable pieces.

Important Decisions

The argument was about whether to sell them.

The case for was straightforward: no customer would ever see two pieces from opposite ends of the run, the drift was invisible in isolation, and forty pieces is a significant amount of material and labour to write off.

The case against was that the archive records each piece as part of a numbered run, and a run implies the pieces belong to each other. Selling them would have made that claim false in a way that could be discovered later by exactly the kind of owner who reads the record carefully.

They were destroyed rather than sold, and rather than sold unnumbered, because an unnumbered channel is a channel and channels get used again.

Quality Standards

The check that now exists is embarrassingly simple: the first, middle, and last pieces of any run are compared against each other before any of them are numbered.

It takes about ten minutes and would have caught this in the first hour. It did not exist because every previous run had been small enough that drift had nowhere to accumulate.

Most quality standards are written the same way — after something specific, in language that only makes sense if you know what happened.