What a Single Reading Can and Cannot Support

Why this matters

A number on a ticket becomes a fact the moment it is written down, and it stays a fact long after the conditions that produced it are gone. Six months later somebody quotes it back: to a customer, to a warranty adjuster, to a lawyer, or to the tech who has to explain why a unit failed two months after a visit that recorded everything as normal.

The reading was probably correct. What was wrong was the size of the claim built on it. A single measurement supports one narrow statement, and the trade routinely stretches it into five or six that it cannot carry. Knowing which is which is not caution. It is the difference between a ticket that protects you and a ticket that gets read aloud.

A sibling card covers how to corroborate a suspicious reading from an independent direction. This one is about the claims themselves: which one a single reading licenses, which ones it does not, and how long it stays worth anything.

The safety exception, before anything else

The rule that overrides everything below: one reading may condemn, one reading may not clear.

A single reading on the dangerous side of a safety threshold gets acted on immediately, without a second opinion, because the cost of a false alarm is a wasted trip and the cost of a missed one is not on the same scale. A carbon monoxide reading in an occupied space, a combustible gas indication, a ground-fault indication, an insulation resistance far below expected: each one is a stop, not a data point to be confirmed at leisure.

The same asymmetry runs the other way and gets ignored. A single reading that says a hazard is absent is not sufficient. The trade has already codified this on the electrical side: absence of voltage is not established by one reading, it is established by proving the tester on a known live source, testing the conductor, and proving the tester again, which is the live-dead-live sequence at NFPA 70E-2021, 120.5 in the edition your employer's electrical safety program adopts. That sequence exists precisely because a single "no voltage" reading can come from a dead meter as easily as from a dead conductor. Every other hazard deserves the same posture even where no standard has written it down.

The one claim a single reading supports

Written out in full, it is this: at that point, at that moment, under those conditions, the quantity read this value, plus or minus the instrument's band.

Four indices are baked into it: place, time, condition, and uncertainty. Strip any one and the claim stops being supported. Most field arguments are about a number that had its indices removed on the way to the ticket.

Six claims it does not support, and what each would need

A trend. "It is getting worse" needs two readings at the same point under the same conditions, separated enough that the difference exceeds the two readings' combined uncertainty. Two readings whose difference sits inside that combined band are one reading taken twice, and reporting them as a trend is the most common way a healthy machine gets condemned.

A rate. "It is losing charge at this rate" needs two readings plus a defensible interval, and the interval has to be long enough that the timing error is small against it. A rate computed across two visits whose exact times nobody wrote down is arithmetic performed on a guess.

A cause. "That is what is causing it" needs a before and an after around one deliberate change, with everything else held. A single reading cannot separate a cause from a coincidence, no matter how well it fits the story.

A comparison to a specification. "This is within range" is licensed only if the specification's stated condition was reproduced. A nameplate value is stated at rated voltage and rated load. An acceptable range is published at a stated ambient, a stated load, sometimes a stated elevation. If you did not reproduce the condition, either reproduce it or write on the ticket that you did not - the honest note is worth more than the comparison was.

A prediction. "It will make it through the season" is supported by no measurement anyone takes. It is a judgment, and if it goes on paper it goes on paper as a judgment with the reasoning attached.

A representative value for the whole system. One point is one point. Stratification in a duct, a tank or a room, and profile effects in a pipe, mean a single location can sit far from the mean, and nothing on the display tells you how far. A representative value needs a traverse, a set of points, or a location someone has already established as representative.

The shelf life of a reading

A reading decays as evidence for three reasons, and they are worth separating because only the first leaves a mark on the machine. The other two leave nothing at all, which is why they are the ones that catch people.

• The measured quantity drifts. Refrigerant leaks, contacts erode, filters load, bearings wear. This is the decay everyone knows about.
• The conditions change. The same healthy system reads differently at a different ambient, load or occupancy. The reading did not go stale, its comparability did.
• Somebody touched the system. A filter changed, a damper moved, a setting adjusted, another trade on site. This is the fastest of the three, the only one that can invalidate a reading in a minute, and the only one nobody logs.

Practical rule: a reading is quotable as a present-state claim only while you can say that nothing has been changed on the system since and that today's conditions match the recorded ones inside the band you care about. Fail either test and it is history.

And history is not worthless. It is the opposite. An old reading is a poor description of today and an excellent baseline for a trend claim, which is the one thing a fresh reading can never be on its own. That inversion is the reason to record conditions at the time: it converts a number with a short life as a verdict into a number with a long life as a comparator.

Worked example: one number, four conclusions on a ticket

What was measured. A tech on a cooling maintenance visit clamps the compressor conductors at an accessible whip outside the disconnect enclosure, a non-contact reading, and records 18.4 A. The nameplate rated-load current is 21.0 A, so the reading is about 88 percent of it. Outdoor ambient that morning was 78 F. Nothing else on the electrical side is recorded.

What went on the ticket. Four statements: compressor amps normal; compressor is in good condition; no refrigerant charge issue; unit will make it through the season.

Working through them.

Amps normal. This is a comparison to a specification whose condition was not reproduced. Rated-load current is stated at rated conditions, and a compressor's current climbs with condensing temperature, so 18.4 A at 78 F ambient says nothing about what it draws at design conditions on a 98 F afternoon. The second reading that would license it: the same measurement taken near design ambient, or the same measurement with the ambient recorded so that a later comparison is legitimate.

Compressor in good condition. A health claim from one electrical quantity. Current is a weak, compressed signal for machine condition, and it is a single point on a machine with mechanical, electrical and refrigerant-side failure modes that do not all show up in amps. The readings that would license it: the rest of the electrical set, the refrigerant-side pressures and temperatures, and the mechanical checks that belong to that equipment.

No charge issue. A conclusion about a different quantity entirely. Compressor current moves with ambient, indoor load and airflow as well as with charge, so a single amp reading is consistent with several charge states. The readings that would license it: the refrigerant-side measurements the manufacturer's procedure calls for, at the stated conditions.

Will make it through the season. A prediction. No reading supports it, and this is the sentence a customer will read back in August.

What was licensed and never written. One sentence: at the compressor conductors, at that time, at 78 F outdoor ambient with the unit in second-stage cooling, current measured 18.4 A against a 21.0 A nameplate rating, plus the clamp's stated accuracy. Nobody would frame that, and it is the only claim the visit actually produced.

The one free field that changes everything. Ambient temperature was already known and took no instrument to record. Written next to the current, it converts a spot number with a shelf life of hours into a baseline that next year's visit can be compared against - which is the trend claim the shop actually wants and cannot get from either visit alone.

The failure mode, concretely. The compressor fails in August. The customer produces a ticket that says the unit would make it through the season and the compressor was in good condition. The shop's defence is that both statements were opinions, which is true and reads badly. Meanwhile the file contains no ambient, so the 18.4 A cannot even be compared to the 20.1 A somebody measures during the failure call, and the one genuinely useful piece of evidence the shop collected is unusable for the only purpose it could have served.

How to verify you got this right

• Read your own ticket back and mark each written statement with the reading that supports it. Anything left unmarked either gets a reading or gets rewritten as a judgment with its reasoning attached.
• Check every comparison for its condition. If you compared a measurement to a rating, ask out loud what condition the rating is stated at and whether you were at it.
• Count your readings against your verbs. Any word implying change over time - rising, falling, worsening, holding - needs two readings and an interval, and if you only have one, the verb comes out.
• Ask what the reading is worth in six months. If the answer is nothing, add the condition fields that would give it a longer life. They are almost always free.
• Apply the asymmetry deliberately. Before you write anything that clears a hazard, ask what a second, independent check would cost. It is nearly always less than being wrong once.

References

• NFPA 70E-2021, 120.5, for the live-dead-live proving sequence, as adopted through your employer's electrical safety program
• Equipment manufacturer documentation for the conditions at which nameplate ratings and acceptable ranges are stated
• See related: The Cost of Trusting a Single Reading; The Baseline Reading You Should Always Take; The Stratification That Fools a Single Reading; How a Reading Changes With Where You Take It