How to Take a Reading Another Tech Can Reproduce
Why this matters
Your shop already has a number for how far apart two techs land on the same measurement. Nobody has ever computed it, so every comparison the shop makes gets judged against a feeling instead. That is why a year-over-year comparison turns into an argument, why "it read 15.2 last spring and 16.1 today" gets called degradation by one tech and noise by another, and why the honest answer, that neither of them knows, never gets said out loud. An instrument's accuracy specification does not answer this. It describes the box. It says nothing about the two hands, two placements and two operating states that produced the two numbers.
The way you find out is to run the same reading twice, with two people, deliberately, and let the spread tell you what your band is. Everything below is built around that one run and the record that makes it possible.
Before either reading
The duplicate run is the point of this article, so notice what it costs: two people, two openings, twice the exposure. Plan it as one opening with both techs present and both readings taken in that opening wherever the quantity allows, rather than two separate entries into the same enclosure.
A clamp reading on a running motor sits behind the energized-work gate at 29 CFR 1910.333(a)(1), which requires de-energizing before work on or near live parts unless that is infeasible and treats a test that can only be performed energized as one of those cases. Instrument, leads and probe tips must all carry an IEC 61010-1 measurement category and voltage rating at or above the point being read, since an under-rated instrument across a line-voltage transient vents at your hands. Wear the electrical protective equipment 29 CFR 1910.335(a) requires for the exposure, keep the free hand outside the enclosure, and stand out of the line of the opening.
If the conductors are bundled and either tech would have to separate them to get a clean jaw, that is no longer a test that can only be done energized. De-energize and lock out under 29 CFR 1910.333(b)(2), or 29 CFR 1926.417 on construction, prove dead with the live-dead-live sequence in NFPA 70E-2021, 120.5 in the edition your employer's electrical safety program adopts, and separate them dead.
The record, filled in
Here is one real reading written out as a record. The right-hand column is the test for whether a field belongs: what would a second tech have to invent if this line were blank.
| Field | This reading | What a second tech guesses without it |
|---|---|---|
| Quantity and point | Line current, black conductor at the load side of the contactor | Which of three conductors, and which side of the contactor |
| Instrument and function | Clamp meter, AC amps, tenths range | Whether it was a clamp or an in-line reading, and the resolution |
| Placement detail | Jaw square to the conductor, conductor centred, no adjacent conductor inside the jaw | Jaw angle and whether a neighbouring conductor was captured |
| Operating state | Compressor running, 11 minutes into the run, single stage, no defrost | The whole load condition, which moves the number more than anything else here |
| Ambient | Outdoor 88 F, indoor 74 F | Whether a change next visit is the equipment or the weather |
| Settling | Reading stable for 20 seconds before recording | Whether the number had arrived |
| Raw value | 15.2 A | Nothing, but a value alone is not a measurement |
| Who and when | Tech initials, date, clock time | Who to ask |
| Anything abnormal | Access panel removed for the reading | Whether the unit was in its normal configuration |
Two of those fields do work most shops skip. Placement detail is what makes the reading a measurement of the same thing next time rather than a measurement of a similar thing. Operating state is the one field that cannot be reconstructed after you drive away, which is why a sibling article puts it first (see References); this one takes it as settled and moves on to what the record is for.
Repeatability and reproducibility are two different spreads
Define them once, because techs use them interchangeably and they answer different questions.
- Repeatability is the spread when the same person takes the same reading on the same setup with the same instrument, minutes apart. It is the smallest spread you will ever see and it flatters the method, because nothing that varies between people got a chance to vary.
- Reproducibility is the spread when a different person, or a different instrument, or a different day is involved, following the same written method. It is always wider, and it is the number your shop actually needs, because your comparisons are almost never same-tech-same-hour.
An instrument's accuracy specification is narrower than either of them and does not substitute for either. It describes the electronics under laboratory conditions. Reproducibility includes placement, settling, operating state and judgment, which in most field measurements dominate the instrument's own contribution.
Running the duplicate
Five paired readings is enough to be useful and few enough that a shop will actually do it. Method:
- Pick one measurement your shop makes constantly and comparisons hang off. Not the exotic one. The one on every ticket.
- Write the record card first, in the shape above, and have both techs work from it.
- Both techs read the same unit in the same opening, in the same operating state, within a couple of minutes of each other.
- The second tech does not see the first number. This is not ceremony. A tech who has seen 15.2 will reproduce 15.2, and you will have measured agreeableness rather than reproducibility.
- Repeat on five different units. Different units, not five readings on one, because you want the spread that includes access differences and configuration differences.
- Record both raw values, unrounded, plus the paired difference.
Setting the band from what you observed
Convert each pair's difference to a percentage of that pair's average, so units of different sizes can sit in one list. Then set the band at the larger of the observed maximum difference and twice the mean difference, and write down which of the two governed. Taking the smaller would give you a band that one of your own observed pairs already exceeds, which is not a band, it is a wish.
The unit of analysis is one measurement type, on one class of equipment, under one written card. A band established for clamp current on running compressors says nothing about surface temperature on the same machines.
Worked example: the run, the outlier, and the field it added
Five units, two techs, blind second reading, current in amps.
| Unit | Tech A | Tech B | Difference | Percent of pair average |
|---|---|---|---|---|
| 1 | 12.4 | 12.6 | 0.2 | 1.6% |
| 2 | 8.1 | 8.5 | 0.4 | 4.8% |
| 3 | 21.7 | 21.4 | 0.3 | 1.4% |
| 4 | 15.2 | 16.3 | 1.1 | 7.0% |
| 5 | 9.8 | 9.9 | 0.1 | 1.0% |
Mean difference is 1.6 plus 4.8 plus 1.4 plus 7.0 plus 1.0, which is 15.8, divided by 5, so 3.16 percent. Twice the mean is 6.3 percent. The observed maximum is 7.0 percent. The maximum is larger, so the band is 7.0 percent and the record says the maximum governed.
Unit 4 is the interesting one and it does not get quietly dropped. Both techs went back to their notes. Tech A read at 4 minutes into the run; Tech B read at 13 minutes. Nothing else differed. On that class of equipment the current climbs as the system loads up, so the pair was never a duplicate at all, it was two different operating states measured correctly.
That produced the actual deliverable of the whole exercise: the card gained a required field, minutes into the run, with a stated minimum of 10 minutes before the reading counts. The other four pairs were all taken past that mark, which is why they cluster.
Now the part that is easy to get wrong. With unit 4's pair excluded, the mean of the remaining four is 1.6 plus 4.8 plus 1.4 plus 1.0, which is 8.8, divided by 4, so 2.2 percent; twice that is 4.4 percent, and the observed maximum among those four is 4.8 percent, so the corrected band is 4.8 percent, maximum governing again. That 4.8 percent band belongs only to readings taken under the corrected card, with the 10-minute minimum enforced. Every reading already in the shop's history was taken under the old card, so a comparison against historical data still uses the 7.0 percent band. Tightening the band on new readings and then benchmarking them against an uncorrected archive would manufacture findings that are not there.
Using it. A unit read 15.2 A last spring and reads 16.1 A today. That is 0.9 A on a pair average of 15.65, or 5.8 percent. Against the historical band of 7.0 percent, that is inside the noise and is not evidence of anything; the honest ticket says so. Take the same unit from 15.2 A to 17.5 A and the change is 2.3 A on an average of 16.35, or 14.1 percent, which is double the band and is a real change worth investigating. Once a full season of readings exists under the corrected card, the same comparison gets judged at 4.8 percent, and the 5.8 percent case flips from noise to a finding.
The failure mode. A shop with no band judges every change against instinct, which in practice means the tech who wants to sell the repair sees degradation and the tech who does not want the callback sees noise. Both are reading the same two numbers. The band is what removes the argument, and it costs one afternoon and ten readings to establish.
How to verify you got this right
- Hand your card to someone who has never taken that reading and have them take it from the card alone, with no coaching. Every question they ask out loud is a missing field. Add it and re-run.
- Check that the second reading was genuinely blind. If the numbers agree far more closely than your instrument's own specification allows, you measured social agreement.
- Re-read your band's governing term. Say which of the maximum and twice-the-mean produced it, on the record, because a band without that note gets recomputed differently by the next person.
- Confirm both readings sat in the same operating state, using the record rather than memory. In this example that single check was the whole finding.
- Re-date the band whenever the card changes. A band and the method that produced it travel together; if the card gains a field, the old band applies to old readings and the new one starts from the change date.
References
- 29 CFR 1910.333(a)(1) and 1910.335(a) - the energized-work gate and the electrical protective equipment required for a live clamp reading
- 29 CFR 1910.333(b)(2) and 29 CFR 1926.417 - electrical lockout in general industry and in construction, for the case where conductors must be separated
- NFPA 70E-2021, 120.5 - the live-dead-live proving sequence, binding through the employer's electrical safety program or an adopting jurisdiction
- IEC 61010-1 - measurement category and voltage rating for the instrument, leads and tips, binding through the product's listing
- See related: How to Take a Reading That Means Something; What Response Time Does to a Reading You Took Too Soon; How to Decide Whether an Instrument Is Still Fit for the Decision