You will see “AQL 2.5” written into almost every led candles quote you receive from China, and almost no one on either side of the table actually knows what it costs in samples. AQL is not a percentage. It is a sampling plan: it tells you how many units to pull, how many defects you can accept before the lot fails, and how those two numbers scale as your order grows. Get the plan wrong and you are either paying for inspection you do not need or shipping a batch with a dead-led rate your returns team will discover later.
The sampling plan is the whole point
AQL 2.5 at general inspection level II maps lot size to a letter code, and the letter code sets your sample size. A 2,000-piece order lands in the 1,201 to 3,200 bracket, which draws 80 units. A 12,000-piece order draws 200. A 50,000-piece order draws 315. The acceptance number at AQL 2.5 then reads off a second table: at 80 units you accept zero, at 200 units you accept two, at 315 units you accept three.
The number people misread is the zero. Zero-acceptance does not mean “the lot must be perfect.” It means the plan is so strict that a single visible defect in the sample fails the whole order, whatever the true defect rate is. At 80 units you are roughly one bad unit in 250, and at 315 units one in 850, before the math flips against you. So the sampling plan and the AQL value are two independent decisions: the plan sets how hard a miss costs, the AQL sets how hard a hit is allowed.
What counts as a defect in a candle
Generic AQL tables do not tell you this, and it is where led candles inspection breaks down. We grade three classes, and the AQL we apply differs by class.
Major defects are functional: dead LED on power-up, flicker that does not cycle through the modes, battery contact that fails to hold, housing cracks, flame tip that sits off-axis. We hold these at AQL 1.5, one level tighter than 2.5. A candle that turns on but flickers at a fixed brightness is a major defect in our grading, not a cosmetic one, because the flicker depth spec is the reason the buyer chose an electronic candle in the first place.
Minor defects are appearance: scratches visible under close inspection, uneven wax coating thickness on wax-coated shells, a 1 mm step in the flame tip mold seam. AQL 2.5. Weighing a scratched candle against a cracked one is where most supplier and buyer arguments on this line actually happen.
Critical defects are safety: exposed battery terminals, batteries that can be pried out by hand pressure, heat from the driver at the switch contact after two hours of continuous run. AQL 0, zero acceptance at every sample size, and a single critical hit fails the lot regardless of the class breakdown.
The math that survives a buyer call
The double-sampling option is the part buyers rarely request and should. A first draw of half the normal sample at a tightened accept line: if it passes, you stop. If it fails, a second draw decides. For a 200-unit single sample that becomes two pulls of 100, and in practice the second pull catches the operator fatigue that inflates minor-defect counts on long shifts. I have seen a lot fail on a single 200-unit draw with three scratched shells, then pass double sampling with the same underlying rate, because the scratches clustered in one station’s output rather than across the whole run.
Two things to check before you sign off. First, the lot definition. If your factory counts a “lot” as one production day and your order ships in three days, you are really buying three sampling plans at the price of one. Second, the defect log. A pass/fail sheet with no per-unit defect coding is useless for the next order. Ask for the coding sheet, not the verdict. AQL 2.5 only works if the defects behind the number are the same every time.