Good Tools Still Lie: A $3,416 Lesson About a Starrett Dial Indicator and the Real Calibration Problem

I look after calibration and tool purchasing in a 45-person job shop. That doesn't make me a metrologist. I don't build uncertainty budgets from first principles, and I would rather ask a certified lab for help than fake my way through calculations. What I can talk about is what happens between a certified tool and a recordable measurement: the stuff that makes good tools lie.

If there is one sentence I want to leave with you, it's this: the shop doesn't trust tools. The shop trusts systems. Tools are just parts of those systems. The most expensive mistake I have watched, and made, is thinking a high-quality dial indicator or a new mechanical caliper is the system.

The rejection that made no sense

Forty-seven brass sleeves went out to a long-time customer in September 2023. Seven days later, twelve came back with one rejection note: internal diameter taper exceeded 0.003 inch.

Our inspection sheet said the same features were within less than 0.0015 inch of nominal. The operator had not changed the offset. The part had not shifted. And the Starrett dial indicator on that setup had just come back from calibration with a green sticker.

So the easy explanation was the tool. I blamed a good Starrett dial indicator because it looked like the only suspect. We even checked the indicator on a surface plate and watched it repeat within 0.0002 inch. I was not dealing with a dead instrument. I was dealing with a hidden bad link somewhere else.

The Starrett dial indicator was fine. The measurement loop was not.

The clue came when the QC tech put a second Starrett dial indicator on the same stand and got identical readings. Then he took the first indicator off the fixture and mounted it on a rigid height stand. It gave the same correct readings all over again. The problem was the fixture itself.

The articulated arm that held the indicator looked secure, but it had a cracked dovetail shoe. Every time the operator locked the arm, the indicator tilted by a few ten-thousandths. On the bench, the indicator behaved. In the fixture, the dial face was convincing but the contact point was not actually staying perpendicular to the part. It was a mechanical lie.

According to ASME B89.1.10, dial indicator calibration includes checks through the full range of the indicator, not just at zero. We had followed that idea for the instrument but ignored the mounting. The indicator could be accurate and still deliver a bad measurement because the loop around it was flexing.

I could call it a cosine error or a fixture compliance problem. The name doesn't matter. The lesson did not appear in the indicator's calibration certificate.

The same bad logic follows other tools

After that job, I started looking at every pass/fail test on the floor the same way. The exact tools are different, but the pattern repeats.

A voltage tester pen is a good example. It is not a lab-grade instrument, but it is a measurement device. I once watched a maintenance tech refuse to open a junction box because his voltage tester pen kept chirping. The pen was doing exactly what it was designed to do: sensing voltage in live cables in the same bundle. It could not tell him whether that voltage was in the cable he was about to touch. If he had treated the chirp as proof, the pen would have created a false positive.

A differential pressure switch taught me the same lesson in a different way. We had a unit set for 0.8 psi differential. The test person used a 0-100 psi reference gauge because it was calibrated and available. The gauge uncertainty was 0.5 percent full scale, so the reference error could be plus or minus 0.5 psi. That is wider than the switch setpoint. The test result was meaningless, but nobody noticed because both instruments had valid calibration stickers.

The same disease even shows up in camera comparisons. When our maintenance supervisor asked me to resolve a Topdon vs Flir thermal camera debate, I had to stop and admit that I am not a thermographer. I can't speak to detector design or software reliability. What I could tell him was that both cameras would produce wrong data if the target was too small for the field of view or the emissivity setting was wrong. The camera is attached to a user who chooses distance, focus, and reflection temperature. That's the measurement system.

The cost was not just the parts

The rejection cost about $3,416 once we counted rework, freight, and extra inspection time. I have to be honest: I wrote down $3,200 first and corrected it after digging through the chargeback. The number did not upset me as much as what the buyer told our sales rep. They did not expect this from us. They chose us because of precision. One bad measurement setup changed how they saw the entire shop.

Detail is a brand extension. When a measurement setup is sloppy, the customer may never see the cracked arm or the wrong gauge. They only see the result. Spending a little more on verified tooling and qualified setups is not an ego purchase. It is brand insurance.

What I changed after the second rejection

I say second rejection because the same pattern showed up again in Q1 2024 on a smaller order. That time I caught it before shipment. This is not a story about one heroic fix. It is a story about a checklist.

I still recommend a Starrett mechanical tool set with caliper S9921 for routine shop-floor measurements. Mechanical tools don't have battery or software surprises, and the S9921 keeps the measuring faces protected when it's stored in a dedicated drawer. But that tool is not the whole measurement. The caliper, the part, the surface, and the person's hand are one loop. If any link in that loop is broken, the instrument will not warn you.

Our pre-run checks are now intentionally simple:

• Has the indicator or caliper been checked on a rigid reference in the same orientation we will use?
• Is the holder or stand included in that check?
• For pass/fail tests like a differential pressure switch, is the reference gauge range appropriate for the setpoint?
• For electrical or thermal work, do we understand what the tool is actually detecting and at what distance?

That's it. No magic brand, no green sticker that fixes a cracked stand, and no reason to stop buying good instruments. The rule is simple: measure the measurement. If you don't, a good Starrett dial indicator will agree with the defect and look perfectly innocent while it does it.