Why Actual Never Equals Setpoint

Why this matters

Every deviation complaint arrives as one number: it is set for this and it is doing that. That single number is never one thing. It is a stack of separate contributions with separate owners, and the reason these calls go badly is that a tech picks the contribution they know how to adjust and spends the visit on it, while the other contributions are still sitting there when they leave. Pull the stack apart on the first visit and you can tell the customer honestly which part of the gap you can close, which part costs money, and which part nobody can close at all.

Before you start reading temperatures around a firing appliance

Put a personal carbon monoxide monitor on your body, not on the toolbag, before the appliance fires, and leave it on you for the whole visit. You are about to spend an hour taking readings while equipment cycles, which is exactly the interval in which a combustion problem announces itself.

Take every reading you can without opening a panel. Where a measurement must be made inside an energized control enclosure, 29 CFR 1910.333(a)(1) permits energized work only where de-energizing would introduce additional or increased hazards or is infeasible due to equipment design or operational limitations, and live troubleshooting is a recognized case; use a meter and leads rated CAT III at or above the circuit voltage and observe the boundaries and protective equipment NFPA 70E-2021 assigns the task. Where the panel can be dead, open the disconnecting means, apply your own lock and tag under 29 CFR 1910.333(b)(2) in general industry or 29 CFR 1926.417 in construction, and prove dead per NFPA 70E-2021, 120.5.

The call

A commercial tenant space, hydronic heat, one thermostat serving the area. The setpoint is 70.0 F. The tenant keeps a thermometer on a filing cabinet that reads 64.0 F, and they have called three times. Two previous visits raised the setpoint, which helped for a few days each time and then stopped helping. The last ticket says "control not holding, recommend replace thermostat."

Six degrees is the complaint. The whole job is deciding what those six degrees are made of.

Establishing the gap is real before explaining it

Two things get checked before anything gets blamed, and both have killed this kind of call before it started.

Compare the customer's instrument against a calibrated reference, at their instrument, at the same moment. Here they read within 0.2 F of each other, so the tenant's thermometer is not the story. That mattered: if their instrument had read 3 F low, the entire remaining investigation would have been chasing an error that did not exist.

Confirm the reading is at steady state, not mid-recovery. The tenant reads their thermometer when they arrive in the morning, which is thirty minutes into a setback recovery. A loop climbing out of setback is supposed to be below setpoint; that is not a deviation, it is a recovery time. Take your own readings at least a full load cycle after any step change, and take them at the time of day the complaint actually describes. Here, mid-afternoon readings showed the same 6.0 F gap, so the complaint survived the check and the recovery timing was ruled out as the cause.

Pulling the six degrees apart

Three contributions, measured one at a time.

Placement, 1.5 F. The thermostat is on a partition wall. Behind that partition runs a warm chase. A calibrated reference taped at the thermostat's sensing point read 1.5 F warmer than the same reference standing in the middle of the occupied area. The controller is not wrong about the temperature; it is right about a place the tenant does not sit. This contribution belongs to whoever mounted the thermostat, and the remedy is physical: relocate the sensor or add a remote sensing element in the occupied zone.

Instrument spread, 0.5 F. The controller displayed 0.5 F above the calibrated reference at the same sensing point. Both instruments are inside their stated tolerance; neither is faulty. This contribution belongs to the instruments themselves and it does not go away, it only gets smaller if somebody buys better instruments. Any honest deviation spec has to allow for it.

Proportional offset, 4.0 F. The controller's own display read 66.0 F against its 70.0 F target, and its output sat steady at 40%. That 4.0 F is not a failure; it is the price of a proportional-only controller with a 10 F throttling range and no manual bias, which generates output only by holding error. Forty percent of a 10 F range is 4.0 F, and the loop must hold that error to keep producing that output. The mechanics of why belong to the proportional-response article; here what matters is that the contribution exists, is calculable, and is the only one of the three that a setting can change.

Add them up: 4.0 plus 1.5 plus 0.5 is 6.0 F, which is the complaint exactly. Nothing is unexplained, and nothing on that list is a defective thermostat.

Why each contribution has a different owner

This is the payoff of separating them, because they do not respond to the same actions.

The 4.0 F offset shrinks if the throttling range is narrowed, and it disappears entirely if reset action is added, which is what integral action is for. Both of those have costs covered in their own articles; neither is free, and narrowing the range too far buys the offset back as oscillation.

The 1.5 F placement error is immune to every setting in the controller. Turning the setpoint up moves the whole system 1.5 F further from where the sensor thinks it is, which is precisely why the previous two visits helped briefly and then stopped: raising a target does not correct a measurement that is reading the wrong place, it just relocates the same error to a higher temperature.

The 0.5 F instrument spread is not correctable at all at this equipment level. It belongs in the deviation specification, not in the repair. An acceptable-deviation band of plus or minus 1.5 F on this loop is not a soft target, it is arithmetic: two field instruments and a sensing location cannot agree closer than that.

What changes the answer

A higher load fraction changes the offset dramatically, and this is the one worth predicting out loud. The offset scales with required output. At 40% output the loop holds 4.0 F of error; on a design day requiring the full output, the same 10 F range and the same zero bias would require 10.0 F of error, putting the controller's own display at 60.0 F and the occupied zone near 58.0 F once placement and instrument spread are added. A tenant who is uncomfortable at 64 F in mild weather is going to be far worse off in a cold snap, and saying so on this visit is the difference between a fixed job and a fourth call.

A controller with a manual bias behaves differently and can hide the same fault. If somebody has set the output at zero error to a nonzero value, the loop can sit dead on setpoint at one particular load and overshoot at every lighter load. A system that is perfect in November and hot in April is showing you a bias tuned for one condition, not a sensor problem.

If the contributions do not sum to the complaint, stop apportioning and start looking for a fourth one. An unexplained remainder is a finding. It usually turns out to be stratification, a second uncontrolled heat or cooling source in the space, or infiltration at one part of the area that the single sensor cannot see.

The failure mode this call is a case study in

Raising the setpoint to chase a placement error is the specific trap, and it is attractive because it works for about a week. It works because the building is not at the same load next week, and any change to a proportional loop looks effective for as long as the load happens to move in the helpful direction.

The cost lands later. Each raise pushes operating temperature closer to whatever limit protects that equipment, eats margin nobody is tracking, and buries the real finding under a setpoint that no longer matches any document. When the limit finally opens, it opens on a cold day months later, and the tech who arrives sees a tripped protective device and no record of why the target is where it is. That tech's correct first step is to establish why the device opened rather than to reset or replace it, and the setpoint history they need to do that has been erased by three undocumented adjustments.

Confirming the split was right

Re-measure the placement contribution at a different load. Move the reference between the sensing point and the occupied area again on a colder day. A placement offset driven by a warm chase gets larger when the chase runs hotter, so a contribution that stays flat across two very different conditions was probably not placement at all.

Confirm the offset by moving the output, not the setpoint. Note the output percentage and the controller's displayed error at two different loads. If the ratio between them tracks the throttling range, the offset explanation is confirmed and the number is predictable at any load. If it does not track, the controller is not doing what its configuration says, and that is a genuine decision-stage finding.

Write all three contributions on the ticket with their owners. The tenant is entitled to know that one part is a mounting location, one part is instrument physics, and one part is a control characteristic with a known price to remove. A single number on a ticket invites a fourth call; a split invites a decision.

References

• 29 CFR 1910.333(a)(1) and (b)(2), OSHA general industry work practices for energized and de-energized electrical work
• 29 CFR 1926.417, OSHA construction lockout and tagging of circuits
• NFPA 70E-2021, 120.5, verifying an electrically safe work condition
• See related: Proportional Response in Plain Terms; What a Setpoint Is and What It Is Not; What Reset and Integral Action Actually Do