Why Your Measurement Data Keeps Failing Review (And It's Not The Instrument)

I review measurement reports from our engineering teams—roughly 400 unique items annually. In Q1 2024 alone, I rejected 12% of first deliveries due to specification discrepancies. My team knows the drill: by the time a report reaches me, the data has been collected, analyzed, and formatted. If it fails, that's a costly redo.

Here's what I kept seeing: a perfectly good Keysight oscilloscope reading, say an MXR608A capturing a 6 GHz clock signal, paired with a report that claimed the signal had unacceptable jitter. The engineer blamed the device under test. The vendor blamed the measurement setup. I blamed... neither.

The surprise wasn't the failed review. It was the root cause. Period.

The Surface Problem: 'My Data Doesn't Match Spec'

You've seen it. You run a measurement on your Infiniium S-Series oscilloscope—let's say you're characterizing a new power supply's ripple with a 34461A multimeter in parallel. The scope says 15 mV p-p. The multimeter says 10 mV RMS. Neither is wrong. But the report shows the ripple is 'out of spec' because someone compared the peak-to-peak value to an RMS limit.

It's tempting to think the instrument is at fault. But the instrument isn't the problem. The problem is how we define 'the measurement.'

The Deeper Reason: Three Hidden Variables

After rejecting the same kind of error across four different projects in 2023, I started tracking what actually caused the mismatch. Three patterns emerged:

1. The 'Precision Gap' Between Spec and Practice

Everyone knows an 8.5-digit multimeter like the 34465A can measure to 0.002% accuracy. But that's under controlled conditions: 23°C, after a 30-minute warm-up, using a specific cable. In reality, your probe adds capacitance, the ambient temperature shifts by 5°C, and your cable loss at 1 GHz is 0.3 dB. That's not a flaw in the instrument. It's a feature of physics.

We specify instruments based on their ideal performance, then get surprised when real-world results drift. The question isn't 'Is the instrument accurate?' It's 'Accurate under which conditions?'

2. The 'Operator Variability' Nobody Tracks

I ran a blind test: same DUT, same test setup, same Keysight spectrum analyzer, three different engineers. One used an average of 10 sweeps. Another used a peak hold over 10 seconds. The third used a max-hold trace. The reported spurious emission level varied by 4 dB. The engineers all swore they 'followed the procedure.' The procedure said 'capture the spectrum and record the highest spur.'

That's the hidden variable: how we interpret the measurement instruction.

3. The 'Old Calibration' Assumption

This was true five years ago when calibration cycles were routinely 12 months. Today, with higher clock speeds and tighter margins, a 12-month cycle is often too long. The 34461A might be 'within calibration' but drifting toward the edge of its spec. We reject reports based on a 0.01% discrepancy that's within both instruments' combined uncertainties—but nobody calculates that combination.

The Cost of Ignoring These Variables

In early 2023, we rejected a batch of 50 power amplifiers because a test report showed 0.5 dB excess gain. The cost: $18,000 for re-testing and re-qualification, plus a 2-week shipping delay. The actual error? The operator used a 1-meter cable instead of the specified 0.5-meter cable, adding 0.3 dB loss and making the DUT look like it had more gain. The vendor said it was 'within industry practice.' We rejected the batch anyway.

That $18,000 mistake could have been avoided if we'd tracked which cable was used. Simple.

A Practical Fix: Reverse-Validate Your Setup

I only believed in setup validation after ignoring it once and eating that $18,000 mistake. Now, every project includes a 15-minute step where someone other than the operator verifies the measurement setup: cable type, probe compensation, instrument settings, and the specific measurement function used. It costs nothing except time. It saves thousands.

Here's the thing: most measurement failures aren't equipment failures. They're setup failures, interpretation failures, or procedure failures. The Keysight oscilloscope is probably fine. The multimeter is probably fine. The question is whether you've accounted for everything between the DUT and the report.

What To Do About It

Three changes that reduced our rejection rate from 12% to under 4% in 2024:

• Document the measurement conditions, not just the result. Include temperature, cable lengths, settings, and the operator's name. If a report fails, you can trace which variable shifted.
• Calculate combined uncertainty. Per industry standards (NIST TN 1900, ISO/IEC 17025), your measurement uncertainty is rarely just the instrument's spec. Include probe, cable, and environment contributions.
• Assume the instrument is right until proven otherwise. When data doesn't match expectations, check the setup first. It's usually a cable, a setting, or a procedure step, not the oscilloscope or multimeter itself.

That's it. The solution isn't a more expensive instrument. It's a more disciplined process.