I'm going to tell you a story that kept me up at night for two months.

I'm a quality director for a consumer electronics brand. We sell Bluetooth earphones. In 2025, we shipped 10,000 units from a factory in Shenzhen. The factory tested every unit. All 10,000 passed.

Three months later, 3,000 of them stopped working. The batteries were dead. Not drained—dead. They wouldn't charge. They wouldn't hold power.

The factory said: "We tested them. They passed."

Our customers said: "They're dead."

I had to find out why.

The First Clue

I flew to Shenzhen. I pulled the test records for the 10,000 units. The records said each battery was charged to 4.2 volts, discharged to 3.0 volts, and charged again. Every unit passed.

But the test records had a pattern. The passing units were not identical. Some had slightly shorter charge times. Some had slightly longer discharge times. The differences were small. A few seconds here, a few seconds there.

I asked the factory: "What's the charge current?"

"1C," they said. "500mA for these batteries."

"Show me."

The Second Clue

I watched the testing line. The batteries were placed in a testing fixture. The fixture connected to the battery, charged it, discharged it, and logged the data.

But there was a problem. The fixture had 20 channels. Each channel tested one battery. The 20 channels were connected to a single power supply. The power supply had a voltage drop when multiple channels were running.

When all 20 channels were active, the voltage at the battery was 4.0 volts. When only 1 channel was active, the voltage at the battery was 4.2 volts.

The test was not consistent. The voltage varied by channel and by load.

The Third Clue

I asked: "When do you run the test?"

"During the day shift," they said. "Eight hours per day."

"Why not the night shift?"

"Because the night shift is for production, not testing."

That was the answer. The daytime test was the most accurate because the line was running at capacity. The night shift test was less accurate because the line was idle.

But that didn't explain the failures.

The Fourth Clue

I looked at the battery manufacturer's spec sheet. The battery was rated for 500 charge cycles at 1C charge and 1C discharge.

But the factory was charging at 1C and discharging at 2C. They were using a faster discharge to speed up the test. The faster discharge was within spec. But it was at the edge of the spec.

The battery was tested at 2C discharge. That's normal for consumer electronics. But the test was short—only 10 minutes. The real-world use was different—hours of discharge at lower rates.

The test wasn't catching the real-world condition.

The Fifth Clue

I started watching individual batteries through the testing process. I marked 100 batteries. I tracked them from the start of the test to the end.

I noticed something. The testing line was intermittent. Sometimes the test took 10 minutes. Sometimes it took 15 minutes. The difference was the software.

The test software had a pause. After charging and before discharging, there was a 5-second pause. The pause was supposed to allow the battery to settle. But the pause was not consistent. Some channels paused for 5 seconds. Some paused for 2 seconds. Some paused for 8.

The pause affected the battery's voltage reading. A battery that settled longer had a higher open-circuit voltage. A battery with a shorter pause had a lower voltage. The test result was inconsistent.

The Sixth Clue

I correlated the pause time with the failures. The batteries with the shortest pause had the most failures. The batteries with the longest pause had the fewest.

The pause was the answer.

The test was designed to pause for 5 seconds. But the software timing was based on the processor clock. The processor clock was slightly fast on some testers and slightly slow on others. The actual pause ranged from 2 to 8 seconds.

A 2-second pause didn't allow the battery to settle. The voltage reading was artificially low. The test passed the battery anyway because it was within the pass/fail window. But the test didn't fully charge the battery. It thought it did. It didn't.

The battery was undercharged. Undercharged batteries have a shorter lifespan. They die faster.

The Full Picture

Here's what happened.

The test was designed to be consistent. But the hardware and software introduced variation. The power supply voltage dropped under load. The software pause was inconsistent. The test conditions varied.

The batteries that passed were not all charged the same. Some were undercharged. The undercharged batteries died faster.

The factory didn't notice because they only looked at the pass/fail result. They didn't look at the test conditions.

The Solution

The solution was a combination of hardware and software fixes.

First, we replaced the power supply with a higher-capacity unit. The voltage drop was eliminated. All 20 channels now received the same voltage.

Second, we updated the test software. The pause was fixed at exactly 5 seconds, using a hardware timer instead of the processor clock.

Third, we added a test condition check. The test would only pass if the charge voltage, discharge current, and pause time were all within spec. If any condition was out of spec, the unit would be flagged for retest.

The failures stopped. The 10,000 units we shipped after the fix had a failure rate of less than 1%.

What I Learned

Here's what I learned from this.

First: Test conditions are more important than test results. The test result said "pass." But the test conditions were inconsistent. A "pass" under inconsistent conditions is not a reliable result.

Second: Small variations matter. A 2-second pause difference is small. But it was enough to cause thousands of failures.

Third: The pass/fail line is not a safety net. The factory only checked if the unit passed. They didn't check the test conditions. They assumed the test was consistent. It wasn't.

Fourth: The factory's quality system was focused on the result, not the process. They had a quality system. It checked the output. It didn't check the input conditions. The input conditions were the problem.

What I'd Tell Every Buyer

If you're buying consumer electronics from China, ask these questions:

Question 1: "What are your test conditions? Not just the test spec. The actual conditions—voltage, current, timing, and pause settings."

If they can't tell you, they haven't checked.

Question 2: "How do you ensure test conditions are consistent across all test stations?"

If they don't have a calibration process, the conditions are not consistent.

Question 3: "Do you log test conditions along with test results?"

If they only log the results, they can't track the conditions that produced them.

Question 4: "What's your hardware calibration schedule?"

If they don't calibrate their test equipment, the conditions will drift over time.

The Aftermath

We're still using the same factory. But we're not using the same test process. The new test process is controlled. The conditions are logged. The failures are history.

The fix cost about $2,000 in hardware and two days of software work.

The problem cost us $150,000 in returns and replacements.

The fix was cheaper.

The Question I Still Have

I still wonder: how many other factories have inconsistent test conditions? How many products pass testing because the test conditions were favorable, but fail because the conditions were not consistent?

I don't know the answer. But now I always check the test conditions, not just the test results.

Factory testing said the batteries were good. Customer testing said they weren't. I spent eight weeks watching the prod

10,000 earphones passed test. 3,000 died in three months. Same battery, same test, same factory. Something was different. I watched the line for eight weeks to find it.

Jangan direproduksi tanpa izin：Blog-Pulviso » The Bluetooth Battery Mystery: 10,000 Units Passed Testing – Then 3,000 Died in 90 Days