What a Plus or Minus Specification Actually Covers

Why this matters

The plus-or-minus figure on a spec sheet is the number every tech quotes and almost nobody has read the boundary of. It is not a statement about your measurement. It is a warranty on the instrument's own indication, under conditions the manufacturer names, with a specific list of things deliberately left outside it. In ordinary field work the excluded terms are usually larger than the covered one, sometimes by a factor of four or more, which means a shop can spend real money upgrading the part of the error that was never the problem.

This card is about the negative space: what the specification refuses to cover, why it refuses, and where the error actually lives once you add the uncovered terms back in.

What the figure is a promise about

Strip it to the core and a published accuracy statement says this: presented with a known input at its terminals, inside the stated conditions, within its calibration interval, the instrument will indicate a value no further from that known input than the stated figure.

Four qualifiers in that sentence do all the work, and each one is a boundary you can step over without noticing.

• At its terminals. The promise begins where the signal enters the instrument, not where the probe touches the world.
• A known input. The promise is about faithfully reporting what arrives, not about that arriving quantity being the one you care about.
• Inside the stated conditions. Ambient temperature band, frequency range, waveform assumptions, orientation, warm-up time, and whatever else the sheet names.
• Within its calibration interval. Some sheets publish different figures for different elapsed periods since calibration, and the longer figure is the honest one to plan around.

Everything below is outside those four qualifiers.

What it excludes, and why the manufacturer had no choice

The accessory is excluded unless separately specified. The clamp jaw, the temperature probe, the pressure transducer, the current transformer, the flow head. The manufacturer cannot know which one you attached, so the accessory carries its own separate published tolerance, usually as a class stated on its own datasheet. For a general-purpose thermocouple probe, that class tolerance is written as the greater of a fixed temperature or a percentage of reading, which means it behaves nothing like the meter's spec and dominates completely at low readings. Two classes are commonly offered, a standard grade and a closer-tolerance grade, and the grade is a purchasing decision that is invisible on the display.

Placement is excluded entirely. No instrument spec has ever contained a term for the probe being in the wrong place. A surface probe on a pipe reads the pipe skin, not the fluid; a thermometer in a return grille reads a mixed stream that may not represent any actual room. A sibling card owns how much a reading moves with where it is taken, and that term is frequently the largest one in the whole stack.

Time since calibration is excluded past the interval. The published figure applies to an instrument inside its interval. An instrument two years past has an unknown accuracy, not a published one, and no amount of careful technique converts one into the other.

Ambient outside the stated band is excluded, and replaced. The sheet gives a temperature coefficient per degree outside the band. A meter that sat on a dashboard through a summer afternoon and got used on arrival is outside its own claim, and the correction is additive on top of the base figure rather than instead of it.

Waveform and frequency outside the stated range are excluded. An averaging AC instrument is specified on a clean sine wave and its error on a distorted waveform is not bounded by the accuracy statement at all. A true-RMS instrument publishes a crest-factor limit and the same applies beyond it.

The pattern behind all of these: a manufacturer can only warrant what it controls. It controls the box. It does not control your probe, your placement, your attic, your waveform or your calendar, so it excludes them and states that it has.

The exclusion that bites hardest: connecting the instrument changes the thing

Every measurement takes something from what it measures, and no accuracy spec covers what your connection did to the system.

A voltmeter with high input impedance draws almost nothing on a real circuit and reads a ghost voltage on a disconnected conductor that is only capacitively coupled, so the excellent spec produces a confident wrong answer. A series current measurement inserts the meter's own burden voltage into the circuit, which on a low-voltage control loop can be enough to change the behaviour you were trying to observe. A pressure gauge with meaningful internal volume, plumbed into a small closed system, lowers the pressure it is reporting. A surface temperature probe with real thermal mass, pressed against a small light part, cools the spot it is reading.

Do not break a live circuit to insert a meter in series. Opening an energized conductor to land it on a test lead creates an arc at the break and puts you across the circuit; 29 CFR 1910.333(a)(1) permits energized work only where de-energizing is infeasible, and inserting a series ammeter almost never qualifies because a clamp accessory measures the same current without breaking anything. Where the series measurement is genuinely required, de-energize and lock out under 29 CFR 1910.333(b)(2) in general industry or 29 CFR 1926.417 in construction, prove dead with the live-dead-live sequence at 120.5 of NFPA 70E-2021 in whichever edition your employer's electrical safety program has adopted, make the connection, then re-energize with the enclosure closed. Instrument, leads and probe tips carry a measurement category and voltage rating at or above the circuit under IEC 61010-1, which binds through the listing mark on the instrument, and 29 CFR 1910.334(c)(2) requires you to inspect the instrument, leads, cables, probes and connectors for external defects before each use.

Worked example: where the error actually lives in one temperature reading

A tech measures a hot surface with a thermocouple-input meter and reports 200 degrees C. A surface at 200 degrees C burns on contact well inside a second, so the probe goes on with a clamp, a magnetic mount or an insulated handle rather than a bare hand, and the reader stays clear of anything that could vent or discharge off that surface while the probe settles. The whole example runs in degrees C because the probe's class tolerance is defined on that scale, and a percent-of-reading term on a temperature scale is defined on the scale it was specified on, not on whatever scale you converted the display to.

Say the meter's published thermocouple-input spec reads plus or minus (0.1 percent of reading + 0.3 degrees C), which is an illustrative stand-in for your own instrument's sheet.

Meter term. 0.1 percent of 200 is 0.2, plus 0.3, for 0.5 degrees C.

Probe term. The probe is sold to the standard grade, whose class tolerance is the greater of 2.2 degrees C or 0.75 percent of reading. At 200 degrees C, 0.75 percent is 1.5, so the fixed 2.2 governs. The greater-of structure is why: read the rule, do not assume the percentage always wins.

Combined, worst case. 0.5 plus 2.2 is 2.7 degrees C. Root-sum-square gives the square root of (0.25 plus 4.84), which is 2.26 degrees C, and the conclusion below is identical either way, which is why the choice of combination does not matter for this question. The sibling card on accuracy and resolution owns when it does.

So where is the error? The meter contributes 0.5 of 2.7, which is about 19 percent. The probe contributes about 81 percent. Now price two upgrades in improvement rather than in money:

• Buy a meter twice as accurate. The meter term halves to 0.25 and the total becomes 2.45 degrees C, an improvement of about 9 percent.
• Buy the closer-tolerance probe grade, whose class is the greater of 1.1 degrees C or 0.4 percent of reading. At 200, 0.4 percent is 0.8, so 1.1 governs. Total becomes 0.5 plus 1.1, or 1.6 degrees C, an improvement of about 41 percent.

The probe upgrade buys more than four times the improvement of the meter upgrade, and it is the one nobody shops for because it has no display.

And the uncovered term dwarfs both. Neither figure says anything about the probe being on the outside of a pipe rather than in the stream. That difference is routinely larger than 2.7 degrees C on an insulated line and it does not appear in any specification, which is the point of this card: the covered error was 2.7 and the total field error is not knowable from the spec sheet at all.

If you are tempted to strip insulation to reach bare pipe, stop. Thermal system insulation installed no later than 1980 is presumed asbestos-containing under 29 CFR 1926.1101 in construction work, with 29 CFR 1910.1001 as the general-industry counterpart, and cutting or tearing it releases fibres that no glove or face shield addresses because the route is inhalation. Use an existing thermowell or test port, or measure at an already-exposed section, and leave intact insulation intact.

The conditions clause is the real specification

Read the conditions before the figure. A tight number valid from 18 to 28 degrees C ambient is a worse instrument for attic and rooftop work than a looser number valid from 0 to 40, and the sheets rarely present it that way.

Three questions get you most of the way on any sheet. What ambient band is this figure quoted over, and what is the coefficient outside it. What accessory was it quoted with, and does that accessory ship with the instrument or get bought separately to a class. Over what elapsed period since calibration does it apply. If a sheet will not answer those, the figure on the front page is a marketing number and should be treated as one.

How to verify you got this right

Take the instrument you rely on most and build the stack for one real measurement you took this month: the meter term at that specific reading, the accessory term from the accessory's own datasheet, and a plain sentence naming the placement assumption you made and cannot quantify. Most techs find the accessory term is the largest one they can put a number on, and that the placement term they cannot put a number on is larger still.

Then check the one thing that costs nothing: whether the instrument was inside its stated ambient band when you used it. An instrument that lives on a truck through a season spends a substantial part of the year outside its own published claim, and correcting for that is a per-degree calculation the manufacturer already printed.

References

• Manufacturer published accuracy statements and the stated ambient band, frequency range, crest-factor limit, warm-up and calibration-period conditions that bound them
• Probe and accessory datasheets for the tolerance class the accessory was sold to, stated separately from the instrument's own figure
• 29 CFR 1910.333(a)(1) energized-work gate; 29 CFR 1910.333(b)(2) and 29 CFR 1926.417 electrical lockout; 29 CFR 1910.334(c)(2) pre-use inspection; 29 CFR 1926.1101 and 29 CFR 1910.1001 for presumed asbestos-containing thermal system insulation
• NFPA 70E-2021, 120.5, adopted through an employer electrical safety program or contract; IEC 61010-1 measurement categories, binding through the instrument's listing
• See related: The Difference Between Accuracy and Resolution; How a Reading Changes With Where You Take It; What Your Instrument Is Actually Measuring