How to Reproduce a Load Combination a Customer Describes

Why this matters

A customer who says "it happens when we have a lot going" has just handed you the test conditions. Almost nobody takes them. The sentence gets written down as "intermittent under load," the tech runs the equipment on its own, everything reads fine, and the shop spends three visits proving that a working system works.

The gap is translation, not effort. The customer describes a scene in their vocabulary, and the test needs a recipe in yours: which devices, in what order, how long apart, under what starting conditions, and for how long. Every piece of that recipe you fail to collect is a reason your test will come back clean on a system that genuinely fails. The steps below are ordered by what you lose when you skip each one, because in practice they get skipped in roughly that order.

What you refuse to reproduce, before anything else

Reproduction is deliberate stress, and some conditions are never reproduced on purpose.

Never defeat, jumper, hold closed or oversize a protective device to keep a combination running. The device operating is the result you came for. Never induce a combination on a combustion path, a shared flue or vent, a pressurized or refrigerant-bearing path, or anything serving a relief or protective function. Do not reproduce a condition that involves water reaching energized equipment.

Cap deliberate trips of a protective device at two. A breaker, fuse or limit that has interrupted a real fault repeatedly can degrade, and past two you are testing the protector instead of the system. Stop immediately, regardless of count, if anything smells hot, discolors, sounds different, or is too hot to hold a hand near. Where a reading has to come off an energized circuit, that is energized work: rated personal protective equipment, a meter whose measurement category suits the circuit, and NFPA 70E-2021, 120.5 live-dead-live proving on anything you intend to touch de-energized afterward, with the de-energize-and-lockout duty for work on electric circuit parts at 29 CFR 1910.333(b)(2). On mechanical or stored energy, isolate under 29 CFR 1910.147.

Translate the nouns, or you will reproduce the wrong scene

What you lose by skipping it: the entire test, because you will run two devices the customer never meant.

Customer nouns are categories, not devices. "The dryer" may be gas or electric, and those are different loads on different systems. "The heater" might be the central system, a portable unit, or a water heater. "Everything" is between three and fifteen things depending on the person.

Do not resolve this by asking better questions from where you are standing. Walk the space with them and have them point. Then write your own list: each device, what it actually is, where it is fed from, and whether it is on the suspect shared node at all. Half the devices a customer names usually are not, and crossing those off is as valuable as finding the ones that are.

Anchor the time, or you will test under different supply conditions

What you lose by skipping it: the starting condition, which is frequently the load-bearing variable.

Get the clock time and the day, not "mornings." Then get what makes that time different, because it is almost never the hour itself. Common real answers: the space is cold after sitting overnight, the utility supply sags at a predictable hour, a neighboring tenant's equipment runs then, the water heater has not recovered yet, or the shift that runs a particular machine starts then.

The reason this ranks so high: a test run at the wrong time can be perfect in every other respect and still return clean, and it returns clean in a way that looks like exoneration rather than like an invalid test.

Get the sequence and the gaps, or you will build the wrong overlap

What you lose by skipping it: the distinction between a start-overlap fault and a steady-state one, which decides what instrument you need.

Ask which came on first and roughly how long before the next. Minutes are fine, precision is not required. What you are separating is whether the failure needs two things starting near each other, or two things running together for a while.

If the customer cannot sequence it, ask what they were doing rather than what was running. "I turn the machine on, then I go start the wash, then I open the front" produces a usable order from someone who could not have answered the direct question.

Get the delay to failure, or you will stop the test early

What you lose by skipping it: the fault itself, by minutes, on a test that was otherwise correct.

"How long after everything is going does it quit?" A range is fine. Then run your test for at least twice that, and never less than 15 minutes on a suspected thermal or cumulative fault. A tech who reproduces the exact combination and holds it for 5 minutes on a fault that needs 9 will write "unable to reproduce" and mean it.

Hunt for the third actor, or you will keep failing to reproduce a pair

What you lose by skipping it: repeatability, on a combination that is actually a trio.

Customers describe two things because two things is what a sentence holds. Ask directly: what else is always on when this happens, including things that are on all the time and therefore invisible to them. Background loads, a piece of equipment that never gets switched off, a circulating pump, a compressor that cycles on its own.

The tell that you are missing an actor is a test that reproduces sometimes with the named pair. Intermittent reproduction of a supposedly deterministic combination almost always means an unnamed third actor is cycling in and out on its own schedule.

Collect the negative cases, or you cannot narrow anything

What you lose by skipping it: the ability to eliminate, which is most of the diagnosis.

This is the single most valuable and least collected piece of information in the whole interview: when did it run the same way and not fail? A day with the same devices and no failure is a controlled experiment the customer already ran for you, and the difference between that day and a failure day is a short list of candidates.

Ask it as a question about their routine, not about the fault. "Is there a day you use all the same stuff and it behaves?" Weekend versus weekday, one shift versus another, one season versus another. Then find the difference.

Write the recipe down, or nobody else can run it

What you lose by skipping it: the handoff, and the demonstration.

The recipe is five lines: starting condition, devices in order with gaps, hold time, what you are measuring and where, and what counts as a pass or fail. Written down, a second tech can run it, the customer can run it while you are elsewhere, and you can demonstrate the fault in front of the person who has been told twice that nothing is wrong. Held in your head, it is gone by Thursday.

Worked example: from a sentence to a recipe to a reproduction

The complaint, verbatim: "It trips when we are using a lot of stuff, usually in the morning when we open."

Nouns translated. Walking the space, the customer points at five devices. Tracing the feeds, three of the five are on the suspect shared node and two are not. The two get crossed off, which is a real result: the customer's mental model of "a lot of stuff" included two devices that cannot participate.

Time anchored. Opening is 6:45 am. What makes it different: the space sits unheated overnight and the first device runs harder from cold. Not the hour, the starting condition.

Sequence and gaps. Device A goes on first, at open. Device B about 2 minutes later. Device C about 5 minutes after B, so 7 minutes after A.

Delay to failure. The customer puts it at 8 to 12 minutes after opening, which is 1 to 5 minutes after C starts.

Third actor. Asked what is always on, the customer names a background unit that cycles by itself and that they had never thought of as something they "use."

Negative case. Sundays they open exactly the same way and it has never tripped. The difference is that device B is not used on Sundays. That makes B a necessary actor, and it does it without a single measurement.

The recipe, written on the ticket. Start from a cold overnight space. Device A on. Device B at plus 2 minutes. Device C at plus 7 minutes. Hold 15 minutes minimum, measuring at the shared node throughout. Fail is a protective-device operation or a deviation past the flag threshold.

First attempt, same day at 10 am. All three devices, correct order, correct gaps, held 20 minutes. Nothing. Every reading unremarkable.

That result is not a failure of the theory, it is the time anchor proving it earns its rank. The only element that could not be reproduced at 10 am is the cold overnight start, and it was skipped precisely because it is the one input that costs a second trip.

Second attempt, next morning at 6:45 am from a genuine cold start. Same three devices, same order, same gaps. It fails 9 minutes after device A, which is 2 minutes after device C started, landing inside the 1-to-5-minute band the customer described. The recipe reproduced the fault on its first correct run.

That gives a demonstration and a mechanism candidate in one: a fault requiring three actors plus a cold start, failing minutes into the overlap rather than at a start, which points at cumulative loading rather than an instantaneous peak. The second attempt cost about 9 minutes of holding, against 20 minutes wasted on the attempt that was missing one condition.

The failure mode this avoids. A tech who runs the same three devices at 10 am, gets nothing, and writes "unable to reproduce, all readings normal" has produced a document that will be used against the customer on the next call. Everything in it is true. The test was invalid, and nothing in the write-up says so. If a reproduction attempt omits a condition you identified in the interview, that omission goes on the ticket in the same sentence as the clean result, every time.

When the customer should run it instead of you

Some recipes cannot be run by a tech at all: the condition happens at 3 am, or once a week, or only on a shift you cannot bill for. Hand the recipe over rather than dropping it.

Give the customer three things: the recipe in their vocabulary, one thing to write down when it happens (the time, and what was running), and the one observation that means stop and call rather than continue, which is anything hot, burnt-smelling, wet or repeated. Two or three logged occurrences will do what a fourth visit would not, and unlike a visit, they happen at the hour the fault does.

References

• NFPA 70E-2021, 120.5, the live-dead-live process for establishing an electrically safe work condition, and its energized-work provisions where a reading cannot be taken de-energized
• 29 CFR 1910.333(b)(2), OSHA electrical safe work practices, for de-energizing and locking or tagging before work on electric circuit parts
• 29 CFR 1910.147, OSHA control of hazardous energy, for isolating stored energy before mechanical service
• Manufacturer service literature for protective-device ratings and published cycle behavior on the equipment involved
• See related: How to Test Two Systems Running at Once; The Fault That Only Exists Under Concurrent Demand; How to Run a Controlled Test to Reproduce an Intermittent