A reading is a claim. The level is 40 percent, the flow is 12 gallons a minute, the temperature is 340 degrees. Your job, staring at a number that looks wrong, is to find one independent thing that agrees or disagrees with that claim. When two sources that should move together don't, one of them is wrong, and now you've got a real problem to chase instead of a hunch.
A reading is a claim. Your job is to find one independent thing that agrees or disagrees with it.
Independent is the operative word. A second transmitter on the same power supply, cable tray, and ground isn't fully independent; it can share the same failure. A tape measure, a sight glass, the motor's current draw, a delivery ticket, or a hand-held meter usually is. Work top to bottom through what follows, find the symptom that matches what you're seeing, and write down what you find as you go. Before you touch anything, note the current reading, the time, and what the process is actually doing, running, idle, filling, heating. Half of these calls get solved just by comparing the number against something you already know to be true.
The reading is frozen, or barely moves while the process clearly changes
Start with the sensing element itself: pull or isolate it and look at the wetted end for coating, scale, or product buildup, and on a pressure or differential-pressure instrument, crack the impulse line's vent or drain and check for free flow, because a plugged impulse line freezes the reading just as effectively as a fouled sensor. Read the raw milliamp value at the transmitter and again at the input card; a signal that's railed or saturated will show you at both ends.
And check whether the controller is actually in manual, or whether the logic itself is holding the value, a first-out capture that grabbed and held the first value that tripped, or a seal-in rung that keeps holding itself on after the trigger clears. A frozen reading is often the logic, not the instrument.
The reading is noisy, jumpy, or bounces around a plausible average
Note whether the jumps line up with a nearby drive, contactor, or pump starting, that's electrical noise coupling into the loop. Verify the shield is landed at one end only and that the drain wire isn't floating or doubled at the other, because a shield or ground problem produces exactly this symptom and it's one of the most common causes I find. Look at the process itself too: splashing, boiling, two-phase flow, an agitator washing near a probe can all produce a genuinely noisy but real signal. And wiggle-test terminations with power down where required; a loose or corroded connection that changes the reading when you move it is your answer.
The reading is smooth and believable, but wrong by a consistent amount
This is calibration drift, and the honest way to check it is against a known reference at two points, not one, zero and span both. Check zero at a known condition, an empty tank, no flow, ambient temperature, because a zero shift from mounting stress or temperature is common and easy to rule in or out. Confirm the range programmed in the transmitter matches the range programmed in the input card or logic; a mismatch there produces a smooth, believable, consistently wrong number that has nothing to do with the sensor's actual health. And clean the sensor and re-read it: if a partial fouling was producing an offset rather than a total block, the offset will shrink.
The reading looks fine but the process behaves wrong, or the other way around
This is the important one, and it's the case that fools me most often, because a steady, believable number is the last thing you think to doubt. The number says 40 percent and holding, but the pump is short-cycling and the outlet's spitting air. Here you stop trusting the instrument and go find a second witness.
Motor current or power is one: a pump moving real product draws real current, but know your pump type. A centrifugal pump dead-headed against a shut valve draws less current, while a positive-displacement pump against a closed discharge spikes both current and pressure, so use current as a moving-or-not clue rather than a flow gauge. A second, independent sensor using a different measurement principle, radar level against a pressure-based level, for instance, is powerful precisely because the two don't share a failure mode. A manual or physical check, taping a tank, a sight glass, catching a bucket and timing it, feeling a pipe for heat, settles more of these than any electronics ever will. And a mass or volume balance, what went in should show up as level or weight, compared against a delivery ticket or a batch total, catches the case where the books simply don't close and only one instrument explains the gap. This shows up constantly on tank levels and flows across water and wastewater work, and on temperature and pressure loops across heavy process plants generally.
The 4-20 mA reality check
Most process loops still run on a 4-20 milliamp current signal, and current loops are stubborn in a useful way, the same current flows everywhere in the loop, so you can read it at any break point and it should match. Read the loop current itself, not just the engineering value on a display; that means breaking the loop in series or using a true DC milliamp clamp or loop calibrator, an ordinary AC clamp meter won't read a DC loop, and the number on a HART communicator is a reported value, not the measured analog current. If the field device says 12 mA and the card reads something else, that's a wiring or card problem, not a process problem.
A signal driven hard toward zero or pinned high is a fault, not a reading, though don't call a valid 3.9 mA a failure. Healthy transmitters legitimately live down to about 3.8 mA and up to about 20.5 mA at range extremes; the recognized fault bands under NAMUR NE43 sit at or below roughly 3.6 mA and at or above roughly 21 mA, and the narrow sliver in between is a deliberate guard band, not a contradiction. Many input cards can flag under-range, over-range, or open-loop faults if you enable it, and it's worth enabling; on older cards without that option, the raw milliamp value is your fault detector either way. I've spent most of a shift chasing a noisy level reading that turned out to be a shield landed at both ends, so if you find one bonded that way, that's a very common source of the noisy-for-no-reason complaint.
A quick decision sequence
Does the reading make physical sense right now? If the tank can't be at 90 percent because you just drained it, the instrument is the problem and you skip straight to the instrument checks. If it's plausible but suspicious, find one independent cross-check, current, a second sensor, a manual read; agreement points at the process and disagreement points at the instrument.
If they disagree, read the raw loop current at both ends: a mismatch between ends is wiring or the card, agreement between ends but a wrong value is the sensor or its calibration. If the loop is electrically healthy, check calibration at two points and inspect the wetted end for fouling, most wrong-but-steady readings die right here. And if it's healthy, calibrated, and clean, believe it. The process really is doing what the number says, and now you have a process problem to solve. That's a good outcome. You stopped chasing the instrument.
Load cells wander through exactly this same logic, drifting smoothly and believably until a batch total won't close. If that's your specific headache, I've written it up on its own in why your batch weights wander even when the recipe never changes.
The trap specific to this one
Deciding the sensor is lying, and then forcing the value or strapping out the trip so the process keeps running, when the reading may be entirely correct and the tank really is at that level or the process really is over temperature. Don't defeat a protective interlock or bypass a safety instrumented function to make a reading behave. Retuning a genuine nuisance trip, adding a sensible filter, or fixing a grounding problem is legitimate engineering; a protective function that keeps tripping should be treated as data about a real condition until you've actually proven otherwise.
Be honest about the limits here too: an intermittent won't show in a ten-minute look, which is why logging beats any single reading, and two sensors can agree while both are wrong if they share a fouling mechanism or a bad common reference, so treat a cross-check as evidence, not proof.
When a reading has bounced between the instrument vendor, the drive vendor, and the controls integrator, and each says it's someone else's box, the fault usually lives in the seams between systems rather than inside any one of them. Working vendor-neutral, with no product line to defend, means I can look at the whole loop, grounding, signal path, calibration, and logic, instead of just my piece of it.
The printable field version of this guide, with the full symptom-to-check reference, is free in the resources library. If a reading's been bounced between vendors and nobody will own it, get in touch, tell me the loop, and I'll help you corner it.