The New Part That Is Now a Variable, Not a Constant

Why this matters

A customer hands you a system with one component that is three days old and tells you, reasonably, that it cannot be that one. If you accept it, you have quietly deleted a suspect from your list before you took a single reading, and you will never go back to it, because nothing later in the day will make you reconsider something you have already agreed is fine. That is how a shop ends up recommending a major replacement on healthy equipment while the actual fault sits in the one part everyone knows is new.

The whole discipline fits in one rule, and the useful way to learn it is to watch the same rule send two nearly identical jobs in opposite directions.

Constant and variable

Borrow the words from anyone who has run a test. A constant is something you have fixed, so a change in the result cannot be attributed to it. A variable is something that could be moving, so it stays in the explanation.

Diagnosis is exactly this. Your candidate list starts as a set of variables, and each verification converts one into a constant. The list shrinks until what remains is the cause. The method only works if a component becomes a constant for the right reason. Convert one on a bad reason and the list still shrinks, it just no longer contains the answer.

"It is new" is a bad reason. It is a claim about the part's age and about a stranger's purchasing decision. Neither of those is a measurement.

The gate

State it precisely, because a loose version of this rule is what gets skipped under pressure:

A component becomes a constant only when all four of these are true: you personally verified it, in circuit rather than on the bench or on paper, against its own stated requirement, under the condition where the fault appears. This is an AND, not an OR. Fail any one and the part stays a variable and stays on the candidate list, regardless of its age, its invoice, or its packaging.

The unit of analysis is the individual component in this installation, not the part number and not the batch. A part family with an excellent reputation tells you nothing about the one in front of you.

"Under the condition where the fault appears" is the clause people drop. A component checked at rest, cold, unloaded, or with the system idle has been checked in the one state where the complaint does not exist. That check produces a pass, and the pass is worthless.

Case A: the gate closes and the part becomes a constant

A circulating component was replaced by the customer after a supplier told them it was the usual cause. The complaint is that delivery at the far end of the system is weak, and it still is.

Run the gate. In circuit, with the system at its normal operating condition and full demand applied rather than idling, the component's delivery is measured against its own published performance at that operating point. It comes in at 96% of rated delivery, which is inside normal variation for a field measurement on a working assembly.

All four conditions are met. Verified personally, in circuit, under the fault condition, against its own requirement. The component becomes a constant, and it is a real constant, earned by a reading rather than by a receipt.

That closure is what makes the rest of the visit fast. Delivery leaving the component is normal and delivery arriving at the far end is not, so the loss is between them, and the search collapses to the path. It turned out to be a restriction in that path, and the customer's part had never been involved at all.

Note what the gate bought here: it did not disagree with the customer. It agreed with them, on evidence, in about a quarter of an hour, and that agreement is now defensible to anyone who asks later.

Case B: same gate, opposite outcome

Same shape of job. The customer replaced a sensing device, chosen off a cross-reference, because the system was behaving as though that reading was wrong. The behavior did not change.

Run the same gate. The device is physically new, it is the correct part family, it is mounted correctly, and its wiring is right. On a static check with the system idle it responds and reports a plausible value, which is precisely where most people stop.

Now apply the third condition. Under the operating condition where the fault appears, the device's reported value is compared against an independent reference measurement taken at the same location at the same moment. It reads 8% high. Its own documentation states an accuracy of 2%, so it is off by four times its own specification, and it is off in the direction that produces the exact misbehavior the customer is complaining about.

The gate does not close. The part stays a variable, and in this case it is the cause. The reason turned out to be mundane: the cross-reference matched the fitting and the signal type but not the range, so the device is accurate in a band this application never operates in.

Two jobs, one rule, opposite verdicts. Neither verdict came from the part's age.

Why the two cases split: the three ways new fails

The third row is the one that catches experienced techs, because verifying the part number feels like verifying the part. It is not. Identity and configuration are two different facts and a part can be perfect on one and wrong on the other.

There is a fourth possibility worth naming separately because it is not a failure of the part at all: the component is genuinely fine and was simply never the cause. That is Case A, and it is the most common outcome by a wide margin. The gate is not a way of doubting customers. It is a way of making the common outcome trustworthy.

What it costs to treat a variable as a constant

The damage is not that the diagnosis gets slower. It is that the diagnosis becomes impossible while still feeling like it is progressing.

Work it through. Say your list of plausible causes for a given complaint has six entries. Verify five of them properly and the sixth is your answer, and you can defend that answer because each elimination has a reading behind it. Now assume the new part is good without testing it, verify four of the remaining five, and the last one standing gets named as the cause. If the true cause was the part you assumed, you have just produced a confident, fully-documented, wrong recommendation. Every step was sound except the one you never took.

That is why this error is expensive out of proportion to the minute it would have taken to avoid. A slow diagnosis costs hours. An eliminated true cause costs the repair, the callback, the second repair, and the customer.

The tell that it has happened to you: you found a cause, replaced it, and the symptom is unchanged. Before you widen the search, go back and put the assumed-good component back on the list.

When you cannot run the gate in place

Sometimes the condition cannot be created at the moment you are standing there. The building is occupied, the weather is wrong, the season is wrong, or the fault needs hours of runtime. You still have options, in rough order of strength:

• Compare against an identical sibling. Many buildings have two of the same thing. Measure both under whatever condition you can create and compare them against each other rather than against a spec. A meaningful difference between two supposedly identical assemblies is evidence even when neither absolute value is conclusive.
• Run the part's own documented check. Most components have a manufacturer-stated verification that does not require the full fault condition. It is weaker than in-circuit testing under load, so record it as a partial result and say so.
• Reinstall the original, briefly, if it is intact and safe to run. If the old part still functions well enough for a short controlled test, swapping it back is the cleanest possible experiment, because it changes exactly one thing.
• Leave logging in place and capture the real event. When the condition is seasonal or long-cycle, instrument it and come back rather than guessing.

The boundary on that third option matters. Do not substitute-to-test on a combustion path, a pressurized or refrigerant-bearing path, or on any device that serves a relief, limit, interlock or other protective function. Those components are not permitted to be a temporary experiment, and a part whose whole job is to stop a hazard is verified against its specification or replaced, never trialed to see what happens.

And when you genuinely cannot close the gate on this visit, write that down as an open variable rather than letting it drift into the constant column between visits. An untested assumption that survives a shift change becomes a fact nobody can trace to its source.

References

• Manufacturer documentation for the component's stated accuracy, rating, range and configuration options
• Trade-standard practice for verifying a component in circuit under its operating condition rather than at rest
• See related: Confirming a Part Is Actually Bad
• See related: How to Verify a Replacement Part Matches Spec, Not Just Fit
• See related: Parts Cannon vs Diagnosis: The Discipline