AQL sampling will tell you a batch is probably fine. It will not tell you that unit #847 has a cold solder joint on the I2C bus that only fails when the enclosure warms up. If you ship connected hardware, "probably fine" across a sample is not the same as "every unit boots and passes self-test," and the gap between those two is where field returns live. Closing it is the job of a functional test jig and 100% end-of-line testing.
What a functional test jig actually is
A functional test fixture (FCT) — often a bed-of-nails — is a custom rig that the assembled board drops into. Spring-loaded pogo pins make contact with test points on the PCB, power it up, and run a script that exercises the real functions: power rails in spec, MCU flashes and boots, each sensor responds, the radio transmits and the RSSI is sane, every GPIO toggles.
It is the difference between a human eyeballing a board and a script that says PASS or FAIL with a logged reason, in 20–40 seconds per unit, on every single unit.
AQL sampling vs 100% EOL testing
These solve different problems and you usually want both.
- AQL inspection is statistical. An inspector pulls, say, 32 units from a 200-unit lot and checks cosmetics, dimensions, labeling. It catches systematic problems — wrong screws, scratched housings, missing labels — efficiently.
- 100% EOL functional testing is exhaustive on the electrical/firmware side. Every unit runs through the jig. It catches the intermittent, per-unit electrical defects sampling will statistically miss: one bad joint, one out-of-tolerance crystal, one mis-flashed device.
A 32-unit sample from 200 simply cannot guarantee unit #847. Only running #847 through the jig can.
Who pays to build the jig
The fixture is non-recurring engineering. Depending on complexity, a custom FCT jig commonly runs from a few hundred to a few thousand dollars. You pay for it, and you should own it — including the test scripts and the right to move the jig to another factory. Treat it like tooling: get it named in writing as your property. A factory that "keeps" your jig holds a bargaining chip you don't want to hand over.
For small runs, the instinct is to skip the jig to save the NRE. That math usually loses. On a few hundred connected units, a 2% field failure rate that an EOL test would have caught becomes RMA shipping, replacement units, and a tanked review score — far more than the jig cost.
Get the test logs
This is the part people forget to ask for. The jig produces data; insist it comes to you:
- A per-unit log with serial number, timestamp, and pass/fail per test step.
- The yield rate for the run and the top failure modes.
Those logs are your independent evidence that 100% testing actually happened, and the failure-mode breakdown tells you what to fix in the next revision.
How this fits a QC plan
EOL testing is one layer. A complete plan stages inspection across the run rather than only checking at the end:
- Pre-production sample signed off before mass production starts.
- In-line check around 20% through the run, so a drifting process is caught early.
- Pre-shipment inspection on 80–100% of units — which is where the FCT jig belongs.
On a 200-unit EU industrial IoT gateway run, that staged approach plus 100% functional testing is what kept defects under 1% — and because the order went direct to the factory rather than through a Hong Kong distributor, it came in roughly 22% cheaper at the same quality bar. Cheaper and lower-defect are not in tension when the testing is real.
The actionable step: before you place the PO, write the EOL test spec — the exact pass/fail criteria per unit — and make the jig and its logs a deliverable in the contract. If running this remotely is daunting, an engineering-led partner like China Sourcing Agents can spec the jig, fold it into a 3-stage QC plan, and send you the per-unit logs so 100% testing is something you can verify, not just hope for.
Top comments (0)