I don't work for you. I don't work for the factory. I work for the buyer. My job is to show up, inspect the goods, and tell the truth.

Over the last twelve months, I did 24 factory visits across Shenzhen and Dongguan. Most orders passed. Some failed. The failures usually had one thing in common: someone assumed.

Here are the cases that stuck with me.

Letter One: The Capacitor That Wasn't

Date: January 12, 2026
Product: Smart lock PCB
Failure: Gate driver chips failing after 48 hours of testing

I walked into the factory on a Tuesday morning. The production line was running. PCB assembly was moving at a good pace. The visual inspection station was checking each board. Everything looked fine.

But I had a rule: always check the incoming material storage.

I walked to the warehouse. The capacitors were stored in a plastic bin under the window. The sun had been hitting that bin for weeks. The capacitors were rated for 85°C operation. The bin temperature was 45°C. The capacitors had been there for three weeks.

Capacitors degrade in high heat. The electrolyte dries out. The capacitance changes. The gate driver chips that depend on those capacitors stop working.

I flagged the issue. The factory manager said: "We've always stored them there."

I said: "Your failure rate is 12%. That's why."

They moved the capacitors to a climate-controlled room. The failure rate dropped to 1.5% within two weeks. The buyer was happy. The factory learned something.

The buyer's mistake: assuming the factory knew how to store components. They assumed. They didn't check.

Letter Two: The Missing Filter

Date: April 2, 2026
Product: Smart lock PCB
Failure: WiFi module intermittent failures

I was doing a final inspection before shipment. I pulled 30 random boards from the batch. I tested each one. One board failed to connect to WiFi. I pulled another. It connected. I pulled a third. It failed.

This was erratic. That's usually a sign of a design or component issue.

I looked at the WiFi module. There was a filter component on the reference design. It was a small ceramic part next to the antenna. It was missing on the boards that failed. It was present on the boards that worked.

I asked the production manager: "Why is this component missing on some boards?"

He said: "It's not on the production BOM."

I said: "It's on the design BOM. It's on the reference design. Why isn't it on the production BOM?"

He said: "The engineer removed it to save cost. He said it was not necessary."

The engineer was wrong. Without the filter, the WiFi module was sensitive to interference. Some modules worked. Some didn't. The variation was due to manufacturing tolerances.

The buyer had approved the cost-reduction without testing it. They approved a change to the BOM without validating the change. They assumed the engineer knew what he was doing.

They were wrong. The filter cost $0.18. The customer complaints cost $18,000.

Letter Three: The Battery Cables

Date: June 12, 2026
Product: Smart lock PCB (battery-powered)
Failure: Battery draining in 48 hours

This one was a head-scratcher. The lock was designed to run for six months on a single charge. In the lab, it ran for six months. In the field, it ran for two days.

I went to the factory. I tested the boards. The idle current was 200 microamps. It was supposed to be 20 microamps. Something was drawing ten times the expected current.

I traced it to the battery connector. There was a small amount of flux residue between the positive and negative pads on the PCB. The flux was slightly conductive. It created a leakage path that was drawing 180 microamps continuously.

The residue was invisible. The inspection station didn't catch it. The cleaning process was inadequate.

I asked the factory manager: "How do you clean the boards after soldering?"

He said: "We use a cleaning solution."

I said: "Show me."

He showed me the cleaning bath. The solution was three weeks old. It was saturated with contaminants. It wasn't cleaning the boards—it was leaving residue on them.

I said: "You need to change the bath every week."

He said: "It's expensive."

I said: "So is replacing 10,000 locks."

They changed the cleaning schedule. The problem stopped. The buyer learned to ask: "What is your cleaning process? How often do you change the solution?" They learned because they lost $30,000 on returns first.

Letter Four: The Sensor Calibration

Date: August 5, 2026
Product: Smart lock PCB (with fingerprint sensor)
Failure: Fingerprint rejection rate of 18%

This was the most expensive failure I saw last year. An order of 15,000 smart locks with fingerprint sensors. The locks were installed in an office building. 18% of the users couldn't unlock the door with their fingerprint.

The problem was calibration. Each sensor needed to be calibrated with a known finger. The factory was calibrating the sensors using a generic calibration. They were doing the same calibration for every sensor. They were assuming the sensors would be consistent.

They weren't. Sensors have variance. The variance is normal. But if you don't calibrate each sensor individually, that variance becomes a failure.

The supplier tried to argue. "It would take too long to calibrate each one."

I said: "Did you calibrate your sensors for the sample that passed testing?"

They said: "Yes."

I said: "Then why wouldn't you calibrate the production units?"

They didn't have an answer. They had assumed that sample calibration was enough. They were wrong.

The buyer had to replace 2,700 units. The cost was $135,000. If they had asked: "How do you calibrate each sensor?" they would have avoided it.

Letter Five: The Software Boot Sequence

Date: October 20, 2026
Product: Smart lock PCB
Failure: Locks resetting themselves randomly

This one was software. The hardware was perfect. The soldering was excellent. The components were correct. The locks still reset themselves. Sometimes once a day. Sometimes three times a day. No pattern.

I went to the factory. I looked at the software. The boot sequence was written in a way that timed out if the memory check took too long. When the memory check took 800 milliseconds, it would pass. When it took 850 milliseconds, it would fail and trigger a reset.

The memory check time depended on the flash memory supplier. Supplier A made flash memory that was slightly slower. Supplier B made flash memory that was slightly faster. The factory used Supplier A for the sample. They used Supplier B for production.

They didn't test the production memory before the production run. They assumed the flash memory was interchangeable.

The software had a hard-coded timeout of 820 milliseconds. Supplier B's flash memory sometimes exceeded it. The locks reset. The users complained.

The fix was simple: change the timeout to 1,000 milliseconds. Two minutes of programming. Two years of customer complaints.

The buyer's mistake: assuming the software was validated for all component variations. It wasn't. They didn't ask: "What happens if a component supplier changes?"

The Pattern

After 24 factory visits, I noticed a pattern. The failures had a common thread. Someone assumed something.

The capacitor storage. The missing filter. The dirty cleaning bath. The generic calibration. The software timeout.

In every case, someone said: "We've always done it this way." Or: "It should be fine." Or: "The specs say it's okay."

In every case, the assumption was wrong.

The 2026 QC Checklist

Here's what I tell my clients now. It's not a long list. But it's the list that works.

ELECTRONICS QC CHECKLIST 2026

What I Tell Every Buyer

I'll tell you the same thing I tell every buyer I work with.

A factory visit is not a check-the-box exercise. It's an investigation. You're looking for assumptions. You're looking for places where the factory said "it should be fine."

If you find three assumptions that haven't been verified, you have a problem. If you find five, you have a disaster waiting to happen.

And the factory will not tell you about the assumptions. They don't know they're making them. They just do things the way they've always done them.

It's your job to find them. Or it's my job to find them for you.

Last Thing

I'm not a complicated person. I'm an inspector. I check things. I find problems. I write them down.

The best clients I have are the ones who read my letters and fix the problems before the next order. The worst clients are the ones who say "that's just how it is" and keep ordering.

The first group doesn't have recalls. The second group keeps me employed.

I inspect electronics for a living. This is what I saw last year. The brands that failed never knew what hit them.

Twelve months, twenty-four factory visits. Most orders passed. A few failed spectacularly. Here's what I learned from the ones that did.

Jangan direproduksi tanpa izin：Blog-Pulviso » A QC Inspector's Letters from Shenzhen: 12 Months, 24 Factory Visits, and the Small Details That Ru