How to Sanity Check a Reading Against Physics

Why this matters

A specification tells you whether a number is acceptable. Physics tells you whether the number is possible. Those are different questions, and the second one is cheaper, faster and catches a class of error the first cannot touch: a lead on the wrong terminal, a decimal in the wrong place, a clamp around two conductors, a sample line drawing room air, a sensor that died an hour ago and is still displaying. Every one of those hands you a value that sits comfortably inside a spec range while being physically impossible for the machine in front of you.

Understand what these checks can and cannot do before you use them. A physics check is one-sided. It can condemn a reading outright, and it can never confirm one. A number that survives all five checks below has not been proven correct; it has only failed to be proven wrong. That asymmetry is what makes the checks worth running first: a condemnation is a definite result you get in seconds.

The order, and why it is this order

The five checks are ordered by how much of the error space each one eliminates per second spent on it, not by how fundamental the physics is. Check 1 costs nothing and kills the most common field error. Check 5 requires a nameplate, some arithmetic and a judgment about what counts as close, so it goes last even though it is the most general.

Before any of these readings

Energized work. Every current and voltage check below happens inside an open enclosure. 29 CFR 1910.333(a)(1) requires live parts to be de-energized before you work on or near them unless de-energizing is infeasible, and its note treats testing that can only be done with the circuit energized as one such case; the gate covers the reading and closes the moment you land, remove or reposition a conductor. Meter, clamp, leads and probe tips must all carry an IEC 61010-1 measurement-category and voltage rating at or above the circuit, because an under-rated instrument across a line-voltage fault vents as an arc at your hands. Use the electrical protective equipment required by 29 CFR 1910.335(a) for that exposure, keep the free hand out of the enclosure, and do not stand square in front of a panel you are about to work in. Repair work means de-energize and lock out under 29 CFR 1910.333(b)(2), or 29 CFR 1926.417 on construction, proving dead with the live-dead-live sequence of NFPA 70E-2021, 120.5. 29 CFR 1910.147 governs the mechanical and stored-energy side, a blocked fan or a pressurized vessel, and expressly excludes this electrical work at 1910.147(a)(1)(ii)(C).

Combustion. Before an appliance fires for any flue reading, a personal carbon monoxide monitor runs in the occupied space. If ambient carbon monoxide climbs toward the 50 ppm 8-hour time-weighted average of 29 CFR 1910.1000 Table Z-1 or the 200 ppm ceiling NIOSH publishes, everyone leaves the building immediately, no switch is touched on the way out, and the space is ventilated from outside before re-entry. A flue probe comes out hot enough to burn through a light glove: handle it by the grip and set it on a non-combustible surface.

Check 1: does the sign point the right way

Ask which direction the physics requires, then look at your number.

Heat moves from hotter to colder without work being done on it. Pressure falls in the direction of flow through a restriction. Current flows into a load, not out of it. A gas passing through a burner loses oxygen, it does not gain any.

What this eliminates: reversed leads, swapped probes, a differential taken in the wrong order, and a sign convention misread. It costs one second and it catches a wrong answer that would otherwise look entirely plausible, because reversing a differential preserves the magnitude and only flips the sign, which is exactly the part people do not check.

Check 2: is the value outside a hard physical limit

Some numbers cannot exist. Learn the handful that apply to your trade and use them as a wall.

• Dry air is about 20.9 percent oxygen, so a flue gas sample cannot read above that. It can approach it under total dilution, but it cannot exceed it.
• Nothing gets colder than its surroundings without something doing work to make it so.
• A useful-output-over-input efficiency cannot exceed 100 percent on the same energy basis.
• A flow cannot exceed what the pipe cross-section passes at a physically sensible velocity.
• A winding cannot have zero resistance and it cannot have infinite resistance while still running.

What this eliminates: a dead or saturated sensor, a decimal error, a unit slip, and a sample path drawing from somewhere other than where you think. This is the check that produced the fastest result in the case below.

Check 3: does it conserve

What goes in comes out. Current into a node equals current out of it. Mass in equals mass out. Energy in equals useful energy out plus losses.

Two cautions that make this check honest. Currents only add arithmetically when the branches share a leg and the readings are taken at the same load state; on a multi-leg supply you cannot sum magnitudes across legs and expect the feeder value, and on a single-phase three-wire supply the neutral carries the difference between the two legs rather than the sum. Take the readings within the same minute under the same load, or the check is comparing two different machines.

What this eliminates: a branch you did not know about, a clamp that captured two conductors, a leak or a dilution path, and a load that switched between readings.

Check 4: does the governing relation close

Where three quantities are locked together by a law, measure all three and see whether they multiply out. Ohm's law is the everyday case: voltage equals current times resistance, so any two of them predict the third.

The caution here is bigger than the check. Ohm's law closes on a resistive element at a steady state, not on a running motor, a transformer or anything with substantial inductance, where the ratio of applied voltage to running current is an impedance that includes effects the simple relation does not model. Applying V divided by I to a running motor and calling the answer a winding resistance is a classic wrong result that survives all the other checks. Use this one only where the relation genuinely holds, and say which element you are applying it to.

What this eliminates: any single wrong member of a triple, when the other two are sound.

Check 5: does it land near a first-principles estimate

Work out roughly what the number ought to be from something independent: the nameplate, the geometry, the design intent. Then ask whether your reading is within a factor of what that estimate suggests, not whether it matches.

Say the nameplate gives a rated input of 5.0 kW at 240 V single phase. Then 5000 divided by 240 is about 20.8 A at full rated load. A measured running current of 18.0 A is 18.0 divided by 20.8, which is about 87 percent of rated, and that is an ordinary place for a loaded machine to sit. A measured 3.2 A on the same machine is 15 percent of rated, which is not a lightly loaded machine, it is a measurement problem or a machine that is not doing its job.

What this eliminates: everything the first four missed, at the cost of the most effort and the most judgment. It is also the only one of the five that gives you a rough expected value rather than just a verdict.

Worked case: three checks, two failures, one clean result

A tech is asked why a burner will not hold its setup and why a panel feels warm. Two readings are in play.

The flue reading. The analyzer reports 22.4 percent oxygen in the flue. Check 2 kills it before check 1 is even needed: 22.4 is above the roughly 20.9 percent oxygen of dry air, and no sample from a fired appliance can exceed the air it came from. Nothing about the burner is in question yet. The instrument or the sensor is the finding, so the tech runs a fresh-air zero, which is the manufacturer's stated procedure for exactly this, and the cell fails to settle at ambient. That analyzer is out of service until it is calibrated. No adjustment is made to the appliance on the strength of a reading from a condemned instrument, because leaning a burner on bad oxygen data is how a carbon monoxide problem gets created.

The panel reading. A feeder is measured at 46.2 A. The three branches the tech can see read 18.0 A, 12.4 A and 9.1 A, all on the same leg, all within the same minute at steady load. Check 3: 18.0 plus 12.4 plus 9.1 is 39.5 A, against a feeder reading of 46.2 A. The gap is 46.2 minus 39.5, which is 6.7 A, and 6.7 divided by 46.2 is about 14.5 percent of the feeder value. That is far too large to be instrument uncertainty on a set of clamp readings taken with one meter on one range.

Conservation does not tell you which candidate is true, only that one of them must be. The tech works the list: a clamp around two conductors would have shown as a low or near-zero reading rather than a shortfall at the feeder; the readings were simultaneous, so a load did not switch; all four conductors are on the same leg, so leg addition is valid. That leaves an unaccounted branch, and there is one, feeding a small load through a separate raceway. It measures 6.6 A.

Now the sum is 18.0 plus 12.4 plus 9.1 plus 6.6, which is 46.1 A, against the 46.2 A feeder reading. The remaining difference is 0.1 A, or 0.1 divided by 46.2, about 0.2 percent, which is comfortably inside the combined uncertainty of five clamp readings. The check now closes, and the warm panel has a real explanation to pursue rather than a mystery.

Note what the closed check did and did not prove. It did not prove any individual current is correct. All five could be biased the same direction by the same instrument and still sum correctly. It proved only that no significant current is unaccounted for, which was the actual question.

How to verify you got this right

Take the last reading you reported and run checks 1, 2 and 3 on it now, out loud, in under a minute. If you cannot state what physical limit would have caught it or what would have to conserve, you did not sanity-check the reading, you accepted it. Then confirm you have not overclaimed: write the finding as "consistent with" rather than "confirms," because a passed physics check is the absence of a contradiction and nothing more. The sibling article on confirming a reading before you trust it covers the positive side, verifying the instrument itself.

References

• 29 CFR 1910.333(a)(1) energized-testing gate; 29 CFR 1910.333(b)(2) and 29 CFR 1926.417 electrical lockout and tagging; 29 CFR 1910.335(a) protective equipment
• 29 CFR 1910.147 mechanical isolation and stored energy, with the electrical carve-out at 1910.147(a)(1)(ii)(C); NFPA 70E-2021, 120.5; IEC 61010-1 measurement categories
• 29 CFR 1910.1000 Table Z-1 carbon monoxide permissible exposure limit; NIOSH ceiling value for carbon monoxide
• Instrument manufacturer's documented fresh-air zero and calibration procedure for electrochemical gas cells
• See related: How to Confirm a Reading Before You Trust It; What Your Instrument Is Actually Measuring; The Difference Between Accuracy and Resolution