A rejected production run is rarely a manufacturing failure. It is usually an approval that was given too early, on a unit that did not represent the final configuration. Four gates, in order, remove almost all of that risk.
Gate 1: the engineering sample
The first unit exists to answer one question — does the specification hold up? Output at the load you specified, temperature under sustained load, the plug mechanism, the port combinations. Nothing cosmetic matters yet. Approve it for electrical performance only, and say so in writing, otherwise the finish gets locked in before the enclosure is final.
Gate 2: the branded sample
This is the unit that carries your logo, colour and finish. Approve it against a physical sample, not a rendering — logo placement is the classic dispute, because a 2mm difference between the artwork proof and the moulded part becomes visible once it is on a retail shelf. Ask for the sample in the packaging as well, not loose, so the accessory set, manual and insert are all approved together.
Gate 3: the packaging proof
Carton artwork, print colours, dieline, barcode placement and the language sheet are approved here. The proof should be a physical printed sheet on the production board stock, because kraft, coated and recycled boards render the same ink differently. Every marketplace programme we run has a rejection story that traces back to a carton approved as a PDF.
Gate 4: the pre-production (golden) sample
Before the bulk run, one unit is built from the actual production line, with production components and the production tooling, and measured against the approved sample. This is the gate that catches the silent changes — a substitute capacitor, a different cable supplier, a slightly different magnet grade. The report from this gate is what the batch is compared to at the end of the run.
What each gate should produce
Not an email saying “looks good”. Each gate should produce a short record: what was measured, what the tolerance is, what was photographed, and who approved it. Twelve lines of text per gate is enough, and it turns a claim six months later into a traceable question instead of an argument.
Where the time actually goes
Sampling adds about two weeks to a programme, and it is the cheapest two weeks in the whole schedule. Everything it prevents — a re-run, a re-print of cartons, a delay in a launch window — costs multiples of that. If a supplier wants to skip sampling for a branded first order, that is the moment to slow down.