Todd pushed the lab door open at 4:15 on a Tuesday, one hand wrapped around a cold Pepsi and the other holding a yellow work order. Our largest account had just placed a rush order: an Anritsu MG3692B signal generator and an Anritsu 37397C vector network analyzer, both headed to a brand-new point-of-presence site that ran almost entirely on Cisco equipment. The customer's RF team wanted both units verified, calibrated, and on a truck within two days.
The bay smelled like coffee and isopropyl alcohol, same as it always does when a ship date is tight. Todd set the Pepsi down on the bench — right next to a calibrated reference, which I made him move — and then asked the question I hear at least once a month.
“Any way we skip the full verification run?” He already knew my answer. “They approved the pricing. Standard cal documents, ship tomorrow, done.”
He meant well. He also wouldn't be the one on the phone when something drifted two weeks later.
I'm the quality/compliance reviewer here. For the past four years, I've been the last person to sign off on every test instrument that leaves our bay — 200+ unique items a year. I don't do the repairs. I do the “is this really what we promised?” part.
In 2022, I implemented the verification protocol we still use today. That was the year a bad batch cost us a $22,000 redo and a delayed launch, and the customer didn't come back. I'm not chasing perfection. I'm chasing repeatability. If a unit ships with a measurement uncertainty the field team doesn't know about, that team will chase the wrong problem — and that's a lot more expensive than a late shipment.
First, the signal generator. The MG3692B is one of our workhorses — solid RF output, reliable leveling, and normally the most cooperative unit in the rack. The bench run started clean: CW output at 1, 5, and 10 GHz, all within ±0.3 dB of the set level. Harmonics looked reasonable. Sweep repeatability was tight. I nearly stamped the certification and moved on.
Then I pushed the test to the upper end of the MG3692B's range at 20 GHz.
The level pulled about 1.2 dB off nominal at the top of the sweep. Still within the published spec, but the trace wasn't as flat as it should've been. On paper, the unit passed.
That's the trap.
Most buyers focus on headline specs — frequency range, price, delivery date — and completely miss the things that actually cause field failures: calibration traceability, connector condition, and whether the verification covered the real operating range. The question everyone asks is “how much?” The question they should ask is “how do you prove it's right?”
At 20 GHz, a 1.2 dB pull isn't just a number on a report. If that generator becomes the reference for a link budget, the whole calculation shifts. The network can test fine in the lab, then mysteriously drop connections in the field.
We re-ran the 20 GHz sweep twice. Same result. Not a hard fail, but close enough that I couldn't sign my name without noting it. We adjusted the reference, extended the warm-up, and reran. The trace settled. The fix cost us four hours, and the ship date moved.
Next, the 37397C vector network analyzer. The instrument itself was in good shape — clean connectors, RF performance within expected tolerances. But the customer, in the name of saving money, had specified a third-party calibration kit instead of the Anritsu kit. Same connector types. Same frequency range. Comparable specs. On paper.
People think premium test gear costs more because the name on the front panel is prestigious. I'd argue it's the other way around: Anritsu can command the price because the kit is traceable, repeatable, and certified. The price is a result of the quality, not the cause of it. When you swap a $400 generic cal kit into a $60,000 measurement setup, you haven't saved $400. You've added an uncertainty nobody sees until something doesn't add up.
We ran a blind comparison: same 37397C, same port, same temperature, same day. Only the cal kit changed. With the Anritsu kit, the S11 trace was clean and repeatable. With the generic kit, the trace wobbled about 0.2 dB across the band. That doesn't sound like much — until you're validating a path with a 1-2 dB link budget margin. A 0.2 dB measurement wobble is the difference between signing off on a healthy circuit and signing off on one that's quietly 2 dB down and will fail at the worst possible moment.
The “budget” choice ended up costing us three hours of re-testing with the correct kit, a renewed certification, and a revised document set. Net loss on that saving: about $1,800 — before we even counted the schedule slip.
Then there was the DuraXV Extreme. One of the customer's field supervisors wanted to use those rugged phones to photograph tower work, upload spectrum analyzer readings, and keep the crew in touch on-site. It's a solid device for that role. The mistake would've been assuming it couldn't affect the measurements.
Here's what happened: we took a production DuraXV Extreme into the lab, placed it next to a portable analyzer, and started sweeping. When the phone's radio went active — just a status ping — the noise floor jumped about 3 dB on the narrow sweep. In the field, that looks exactly like a bad PIM connection or loose RF path. The crew would've spent a full day hunting a ghost fault that was actually a phone in someone's pocket.
We caught it because the 2022 protocol added a “co-located radiating devices” screening step. It takes ten minutes. Ten minutes that can save an entire site visit. I know it sounds kinda obsessive. It is. Obsessive is cheaper than a false diagnosis.
Before I get to the phone call, one thing I keep seeing in buyer search logs: “anritsu vs cisco.” They aren't competitors. Cisco makes the switches and routers that move traffic; Anritsu makes the instruments that prove those paths are healthy. In this order, the MG3692B fed a 10 GbE backhaul test through a pair of Cisco switches, and the 37397C pulled the measurements at the far end. You need both. Treating them as alternatives is like comparing a wrench to a torque spec.
The same logic applies to the paperwork. Per FTC advertising guidelines, every claim we make about these units — and every claim the customer repeats to their own leadership — has to be truthful, not misleading, and backed by evidence. That's why we don't put “100% accurate” anywhere on a datasheet. That's also why every verification run generates a record, and why we keep those records for years. That's not paranoia. That's traceability.
And traceability is where efficiency actually wins. We moved our verification logs to a digital workflow in 2022. The paper binders served us fine for a long time, but they didn't tell us that a unit had drifted during a previous calibration — we had to go looking for it. Now the history is one query away. When the MG3692B anomaly appeared, I pulled the unit's full history in minutes and found a similar pattern in an earlier calibration. That's the efficiency that matters: not moving faster, but finding the problem before the customer does.
We shipped the order two days late. The customer's project lead was not pleased. His exact words, I'm told, were a little more colorful than I'll repeat here.
Then he read the verification notes. Saw the 20 GHz finding, the cal kit replacement, the DuraXV Extreme screening record. The tone changed.
“Wait,” he said. “You actually caught these before shipping?”
That installation went live two days later. The 37397C confirmed every backhaul path through the Cisco gear within expected limits, and the field crew had zero ghost-PIM alarms that week. The supervisor later told me the DuraXV Extreme units are now scheduled for RF screening on every new site, and the company standardized on manufacturer-certified calibration kits going forward.
About a month later, they sent a renewal order. Upgraded verification package, no generic cal kits, and a request for us to train two of their techs on our screening process. I'd call that a win. Not because we were late, obviously. Because the process did its job. The customer saw verification as overhead — until it saved them from chasing a fault that never existed.
If you're shopping for test equipment, here's the unfiltered version:
The certification documents for this order went out by First-Class Mail, by the way. According to USPS pricing effective January 2025, that's $0.73 per letter. It's still the cheapest part of the entire transaction.
Efficiency isn't the same as speed. Speed is shipping things fast. Efficiency is shipping things right the first time, with the proof to back it up. That's what I review for. That's what I'd recommend anyone else review for.
Simple.