What Calibration Actually Establishes

Why this matters

Most shops treat a calibration certificate as a pass mark: it came back, the sticker is current, the instrument is good. What the document actually contains is much narrower and much more useful than that, and the gap between the two readings of it is where a shop quietly loses the ability to defend its own numbers.

A certificate is a historical, point-by-point, condition-bound statement about the instrument alone. It says what the instrument indicated, at a handful of specific values, on one day, in a laboratory, against a reference of stated uncertainty, with the lab's own accessories attached. Every step from there to "so my reading last Tuesday was good" is an assumption you are making, not something the certificate said. This card is about reading the document itself, and specifically about three fields that decide what it means and that almost nobody reads: the calibration points, the laboratory's own uncertainty, and the decision rule.

The companion card on calibration scheduling owns what to do with an out-of-tolerance result, how intervals flex on as-found history, and how to bound the review set of past jobs. This one stops at what the paper establishes.

Before the instrument goes back on the truck

A current certificate does not discharge the pre-use duty. 29 CFR 1910.334(c)(2) requires the instrument, leads, cables, probes and connectors to be visually inspected for external defects and damage before each use, and a unit that was fine in the lab can arrive with a cracked lead from shipping. Where the instrument is used to establish an absence of voltage, prove it on a known live source before and after the test as the live-dead-live sequence at 120.5 of NFPA 70E-2021 requires, in whichever edition your employer's electrical safety program has adopted, because that check catches a failure that occurred after the calibration and no certificate can. Confirm the instrument's measurement category and voltage rating under IEC 61010-1, which binds through the listing mark on the instrument, still suit the circuits it is going back to.

The fields, in the order that decides what the paper means

As-found and as-left. How it read on arrival and how it read on departure. Identical values mean no adjustment was made, which is the outcome that earns confidence. Different values mean it was adjusted, and everything measured with it since the previous calibration was taken on an instrument now known to have been off.

The calibration points. A list of specific values, not a range. The certificate speaks about those points and is silent everywhere else.

The tolerance applied. The limits the lab judged against. Confirm they are the ones you need, because a lab given no instruction uses the manufacturer's published tolerance, which may be looser than the tolerance your work actually requires.

The laboratory's own measurement uncertainty. Every calibration is itself a measurement, and it carries an uncertainty, normally reported at a stated coverage. This number is what makes a pass or fail meaningful, and it is the field most often skipped.

The decision rule. How the lab converted its measured error plus its own uncertainty into the word "pass." ISO/IEC 17025:2017 requires a laboratory reporting a statement of conformity to document the decision rule it applied; that standard binds the laboratory through its accreditation, and binds you only where your own contract or a customer specification calls for an accredited calibration.

The ambient conditions and the traceability statement. The lab temperature and humidity the work was done under, and the chain by which the reference standard traces to a national measurement standard.

The worked certificate

Read as a filled-in artifact. The figures are illustrative stand-ins for whatever your own paperwork carries, and the reasoning is the transferable part.

Work the ratio first. The instrument's tolerance is 1.0 units and the lab's own uncertainty is 0.4 units, so the ratio between them is 2.5 to 1. The long-standing convention is that a calibration should have at least a 4-to-1 ratio between the tolerance being judged and the uncertainty of the judging, and where a customer specification names ANSI/NCSL Z540.3 that ratio, or an equivalent limit on false-accept risk, becomes a contractual requirement rather than a convention. At 2.5 to 1 this calibration is worse than the convention, which does not make it useless, but it does make the next field load-bearing.

Now read the decision rule against the numbers. Under simple acceptance the lab compares the measured error to the tolerance and ignores its own uncertainty: 0.85 is less than 1.0, so it passes. But the lab's own uncertainty is 0.4, so the instrument's true error at that point sits somewhere between 0.45 and 1.25 units. Part of that band is outside the tolerance the certificate just declared it inside.

Had the lab guard-banded instead, accepting only where the measured error is inside the tolerance less its own uncertainty, the acceptance limit would have been 1.0 minus 0.4, which is 0.6 units. The same measured 0.85 would then have been reported as a fail.

So the identical measurement is a pass or a fail depending on a field most people never read, and the word on the front of the certificate is not, on its own, a statement about the instrument. A pass under simple acceptance with a thin ratio means "probably in tolerance," and a shop doing work where being wrong matters should be specifying a guard-banded decision rule when it books the calibration, not discovering the rule afterwards.

Note what this changes about the as-found reading too. This instrument came back with an 0.85 error against a 1.0 tolerance: it passed, it was not adjusted, and it is sitting at 85 percent of its allowed error with a whole interval ahead of it. That is a very different situation from an instrument that came back at 0.05, and both print as "Pass."

The gaps between the calibration points

The certificate above establishes behaviour at 10, 50, 100 and 200 units. It says nothing whatsoever about 5 units, or 30, or 150.

For most instruments that is a reasonable gap to interpolate across, because the dominant error terms vary smoothly. Two situations where it is not, and both are common in field work:

• Working below the lowest calibrated point. Errors that are fixed in absolute terms, rather than proportional to reading, dominate at the bottom of a range. An instrument calibrated no lower than 10 units and used routinely at 3 has been characterized where it is good and used where it is weakest. Ask for a calibration point near your actual working value when you book the work, which costs little and is the single most useful instruction a shop can give a lab.
• Working past the highest calibrated point. Nothing above the top point was measured at all, and extrapolation past the end of a data set is not interpolation. Treat readings above the highest point as uncharacterized.

The general habit: before booking, write down the two or three values your shop actually makes decisions at, and ask for points there. A default point list is chosen for the lab's convenience, not yours.

What the certificate says about your accessory and your attic

Two extensions people make automatically, and neither is in the document.

The accessory. Unless it is listed on the certificate by its own identity, it was not calibrated. A meter calibrated at its terminals with the lab's leads says nothing about your clamp, your probe or your transducer, each of which carries its own separate tolerance from its own datasheet. Where the accessory is the dominant error term, and it usually is, a certificate on the meter alone characterizes the smaller half of your stack. Send the instrument and its accessory together as a system where the lab supports it, and where it does not, keep the accessory's own class documentation with the certificate.

The ambient. That calibration was performed at 23 degrees C in a stable room. Your reading was taken wherever the equipment lives. The manufacturer publishes a coefficient per degree outside the instrument's stated band, and outside that band the published tolerance the lab judged against is not the tolerance you are operating under. An instrument that passed at 23 degrees C is not thereby in tolerance on a hot roof, and the paper never claimed it was.

How to verify you got this right

Take your most recent certificate and answer four questions from the document without guessing: what points was it calibrated at, what was the lab's own uncertainty, what decision rule produced the pass, and was the instrument adjusted. If any of the four is not on the paper, call the lab and ask for a certificate that carries it, because a document missing those fields cannot support the conclusion your shop is drawing from it.

Then do the arithmetic this card just walked: divide the applied tolerance by the lab's uncertainty. If that ratio is under 4 to 1 and the decision rule is simple acceptance, you are relying on a pass that carries real false-accept risk, and the fix is a sentence in your next calibration order rather than a different instrument.

References

• ISO/IEC 17025:2017, which requires a laboratory reporting a statement of conformity to document the applied decision rule and to report measurement uncertainty; binds the laboratory through its accreditation and binds you through your own contract or customer specification
• ANSI/NCSL Z540.3, in the edition a customer specification names, for test uncertainty ratio and false-accept risk requirements where a contract invokes it
• 29 CFR 1910.334(c)(2), pre-use visual inspection of test instruments, leads, cables, probes and connectors; NFPA 70E-2021, 120.5, adopted through an employer electrical safety program; IEC 61010-1 measurement categories, binding through the instrument's listing
• Manufacturer published tolerances, calibration point recommendations and ambient temperature coefficients
• See related: The Calibration Schedule Worth Keeping, which owns intervals, as-found history and the review set; What Traceability Means and Why It Matters to You