The end-of-line tester is the last thing between a workmanship defect and a customer site. Its coverage determines which defects escape, and its limits determine how many good units are rejected on the way.
Both are design decisions rather than settings, and both are usually made too late to be made well.
Coverage follows failure modes, not convenience
The sequence should be built from what can go wrong in assembly rather than from what is easy to measure. A tester that confirms the unit powers up and communicates will pass a charger with a mis-crimped earth conductor.
Deriving coverage from the process failure analysis, rather than from a generic functional list, is what makes the tester a control rather than a formality.
Identity first
Capturing the serial number before any measurement means every subsequent reading is attributable, including readings from units that fail. Testers that record identity at the end lose the data for exactly the units that most need investigating.
This ordering costs nothing and is the difference between a yield problem you can analyse and one you can only count.
Setting limits
Limits come from characterising the product across its expected variation, not from the specification. A limit set at the specification boundary will fail units that are within tolerance, because measurement uncertainty and normal spread both exist.
Too tight generates false failures, which erode confidence in the tester and create pressure to override it. Too loose passes marginal units into the field. Both failure modes are common and both originate in limits set without characterisation data.
Golden units
A reference unit, independently verified and run periodically, proves the station is still measuring correctly. Without one, a drifting tester and a genuinely failing batch look identical, and the usual conclusion is that the product is at fault.
Golden units need their own control: stored properly, re-verified on a schedule, and replaced before they age into being a source of error themselves.
Safe test architecture
Any station handling mains potential needs interlocked access, defined isolation and a documented safe state on abort. The abort path should be designed first, because a tester that cannot fail safely will eventually be bypassed under schedule pressure.
Fixture wear is the dominant maintenance item. Degrading contacts produce intermittent failures that look like product defects, which is why fixture verification belongs in the maintenance schedule.
Records make the tester useful later
Each result stored against serial number, with firmware version, configuration, station, operator and timestamp, is what lets you scope a field problem to a date range or a station rather than to everything built that quarter.
This is also the record that answers whether a returned unit left the factory in a known-good state, which is the question every warranty dispute turns on.
Retest rules
Unlimited retesting converts a failing product into a passing record. How many retests are permitted, and what happens to a unit that fails repeatedly, must be defined before the line is under pressure rather than during.
A failure code taxonomy turns test data into a defect Pareto. Without codes, you know your yield but not your problem.