Why My Measurements Were Wrong: A $3,200 Lesson in Test Equipment (It Wasn’t the Instruments)

What looked like a routine project

In March 2024, I signed off on a test plan that looked bulletproof. A review publisher had asked our lab to evaluate seven blood pressure monitors for its annual buyer’s guide. The feature would name one of them the best blood pressure monitor for the next twelve months. Our part was small: verify charging behavior, measure standby power, check basic electrical stability. No RF. No high voltage. No exotic environmental stuff. Or so I told myself at the time.

Two days later, I was staring at 214 readings that contradicted each other. Two units from the same batch disagreed by more than 2:1 in charging current. One monitor drew 0.47 A in one test and 1.29 A in what should have been an identical run twenty minutes later. The client stopped answering my messages. We pulled the report and re-ran the whole electrical validation. The rework and schedule damage came to roughly $3,200 before we counted the credibility hit.

From the outside, a well-equipped lab looks trustworthy. Calibrated instruments sit in rows. Certificates hang on the wall. The reality is that most measurement problems I’ve run into have nothing to do with instrument accuracy and everything to do with how the instruments were connected. The gear was fine. The method was broken. This is the story of how that happened and the checklist I use now to keep it from happening again.

Mistake one: a solid meter in the wrong spot

Our original plan for the charging-current measurement was simple. The monitors charged over USB-C, and we wanted to log current across a full charge cycle. I reached for the handheld meter we use for everything—the one most people are thinking of when they search for a “multimeter 117.” It’s a good meter. Its calibration was current. That made the problem worse, because I stopped thinking.

To measure current with a multimeter, you insert it in series with the load. The meter reads current by adding a small shunt resistor and measuring the voltage across it. That shunt creates burden voltage. With a passive load, you can calculate the error and move on. But a USB-C charger is not a passive load. When a device negotiates charging, its power-management circuitry watches the voltage on the bus and decides how much current to request. If the meter’s shunt drops the voltage by even a few tenths of a volt, the device sees a weak source and backs off. The meter shows you the current that exists after the meter itself has changed the device’s behavior.

So the multimeter wasn’t lying. It was reading the situation it created. We spent an entire morning blaming the monitors.

Mistake two: a human being asked to log USB power delivery

We also needed to see how each monitor behaved in the later stage of charging, when the charge current tapers and the port starts stepping through different voltage states. A second mistake surfaced there. The test procedure required someone to watch the USB power delivery while recording list entries by hand—negotiated voltage, current, port state—into a spreadsheet. A junior engineer got the job. It looked harmless. It wasn’t.

USB power delivery transitions happen in milliseconds. By the time a human notices a new value and writes it down, the port has already moved on. When we compared the handwritten list against the oscilloscope capture, dozens of rows didn’t match the real event stream. The engineer did nothing wrong. No human can reliably transcribe an automated negotiation that moves faster than a stopwatch click.

Mistake three: trusting the front panel instead of the device

The third mistake is the one I’m least proud of, because I knew better. We used a Keysight E3648A bench supply to stand in for the battery during a stability test that required repeated cuff inflations. The E3648A is a solid dual-output supply, and it was doing exactly what it should. The problem was that we connected it to the monitor’s board with long test leads and then trusted the front panel.

A bench supply regulates the voltage at its own output terminals. It cannot compensate for voltage dropped across your wiring unless you use remote sensing or check the voltage at the device itself. When the cuff pump pulled sharp current spikes, the monitor’s input voltage sagged even though the front panel never moved. The unit was browning out mid-test. I suspected the E3648A first. It was innocent. The wires were the problem.

The rental that went right

At least one decision held up. Because some of the monitors transmitted readings to a phone over Bluetooth, we had to characterize the radio output—not for the ranking, but to rule out transmitters that were badly out of spec. Buying a dedicated RF power meter for a few days of work didn’t make sense, so we rented one. The Keysight RF power meter rental turned out to be the least painful part of the whole project. It arrived with a NIST-traceable calibration certificate and a matched sensor, and it produced clean, repeatable data in an afternoon. The rental cost a few hundred dollars—less than 3% of what owning that instrument would have cost us.

The rental’s success made the rest of the week sting more. We had good tools, some bought and some rented. The instruments were not the issue.

The real problem wasn’t any instrument

Here’s the uncomfortable pattern: every piece of test equipment did what it was supposed to do. The multimeter read the current through its own shunt—including the effect of that shunt. The E3648A delivered and displayed the voltage at its output terminals, not at the monitor’s input. The oscilloscope captured the waveforms our probes were connected to, plus whatever the thin leads added. Even the RF meter only worked because we connected it correctly.

People assume that a calibrated instrument plus a clear screen equals the truth. In measurement, all you get is a number. The instrument faithfully transforms whatever physical quantity appears across its terminals into digits. If the terminals see an altered circuit because you inserted a shunt into a negotiation line, or a sagging voltage because your leads are too thin, the digits will be accurate and useless in the same sentence. You’d think I would have learned that in my first year, back in 2017, when I made a similar mistake on a smaller project. Apparently I needed a $3,200 reminder.

What that week actually cost

Let me be specific about the $3,200. It was roughly forty hours of engineer time spent re-running tests, with most of those hours coming from me and my lead technician. It was another $350 for extra time on rented gear while we waited for a logging analyzer to come back from calibration. It was $300 in overnight shipping for replacement monitors, because by then the original units had been handled more than a clinical trial should allow. And it was $550 to expedite that calibration instead of waiting the normal two weeks. I stopped counting once the total crossed $3,200.

The indirect costs were larger. The publisher’s “best blood pressure monitor” feature ran five weeks late, and without our data in it. A competing publication got to the topic first. One monitor that our broken setup had made look electrically unstable was left out of the rerun because the calendar simply didn’t allow it. The lab’s error quietly changed which product got covered. That’s the part that still bothers me. People trust headlines like that when they’re deciding what to buy for a parent with high blood pressure. If we had shipped flawed data, a bad measurement could have steered a health decision.

The scrubbed-down checklist we use now

After the third attempt finally produced clean results, I wrote the checklist that now lives on every bench in our lab. It’s not a work of genius. It’s a scar list. Most of it is boring, which is the point.

1. Trace the full measurement path before you start. Don’t just ask whether the instrument is calibrated. Ask what the instrument is physically inserted into, and what effect inserting it has on the device. If you’re measuring current, check the burden voltage of the meter or shunt at the expected current level.
2. No human transcription of event-driven behavior. If you need to capture USB power delivery while recording list values, use an analyzer or a data logger. A spreadsheet is for reviewing the log after the fact, not for creating the log in real time.
3. Verify voltage at the device, not at the display. The Keysight E3648A we used was reading correctly. Its output terminals, however, were several ohms away from the monitor’s input because of our leads. Use remote sensing if your supply supports it. If it doesn’t, clamp a second voltmeter directly onto the device’s input pins and watch that value.
4. Respect cabling when current draw changes quickly. Thin leads move. Connections oxidize. A crimped connection can look fine and still drop half a volt under the kind of load a miniature pump pulls.
5. Run a sanity check with a known load before testing real devices. A power resistor or a dummy load will tell you if your current measurement is in the right universe. If you can’t reproduce a resistor’s expected value, don’t point fingers at a new product design.
6. Apply the same discipline to rented gear. A rental is not automatically trustworthy because it came with a calibration sticker. Check accessories, run a zero, confirm the correct calibration factor. The Keysight RF power meter rental we used was perfect, but only because we read its setup guide and zeroed the sensor before the test.

None of this means you shouldn’t trust test equipment. I trust ours more now, because we stopped trusting our assumptions. I still use a handheld multimeter for continuity checks and basic voltage troubleshooting, and it’s the right tool for that. It’s just not the right tool for logging current in the middle of a USB power delivery session. There is no universal “best” instrument. There is only the right instrument for a specific measurement, connected with respect for the circuit you’re measuring.

As for the roundup: we eventually delivered data the publisher could use. The feature ran late, the monitor that should have won the spotlight lost it, and I learned more from that failure than from a year of clean test reports. The instrument just reads the result. The setup is the product.