How to Test Two Systems Running at Once

Why this matters

Every test most techs run is a single-actor test. One appliance, one circuit, one fixture, energized on its own while everything else in the building sits idle. That test is blind by construction to an entire family of faults, and it is blind in the most expensive way: it returns clean readings, so the tech leaves with documented proof that nothing is wrong, and the customer gets the same failure that evening when the second thing comes on.

Concurrency faults are not rare. Anything that shares a circuit, a supply line, a vent, a drain, a neutral, a control bus, or a fixed capacity somewhere upstream can fail only in company. The method below is what turns "cannot reproduce" into a measurement, and it takes about a quarter of an hour once you know where the shared node is.

The concurrency you must never induce

Running two things at once on purpose is a deliberate stress. Some paths do not get stressed on purpose, ever.

Do not induce concurrency on a combustion path, a pressurized or refrigerant-bearing path, a flue or vent shared between appliances, or anything serving a relief or protective function. Do not defeat, bypass, jumper or hold closed a protective device to force two loads to coexist: the protector operating is the data, and removing it converts a diagnostic test into an ignition or rupture risk.

Where the shared node is a vent or a drain, the concurrency test is an observation of normal use, not an induced one, and combustion appliances sharing a vent get a combustion-safety check rather than a stress test.

For electrical work, the default is to de-energize, lock and tag, and verify dead before contact under 29 CFR 1910.333(b)(2), proving dead with the live-dead-live sequence in NFPA 70E-2021, 120.5. A concurrent-load test needs readings taken while energized, which is energized work: it proceeds only when de-energizing would itself introduce a greater hazard or make the test infeasible, with rated personal protective equipment and a meter whose measurement category matches the circuit. On mechanical or stored-energy work, isolate under 29 CFR 1910.147 before hands go inside.

Step 1: State the concurrency as a testable pair with an overlap type

Write it as one sentence: which two actors, and what kind of overlap. "The fault appears when the well pump and the water heater element are both energized in steady state" is testable. "It acts up when a lot is running" is not.

There are two overlap types and they produce different faults, so name yours before you plan the test.

• Start overlap: the two events collide only during the inrush or fill of one of them, a window of a few seconds. Faults here are voltage sag, pressure collapse, and control dropout.
• Steady-state overlap: both actors run together for minutes. Faults here are thermal, cumulative, and resistive, and they need dwell time to show.

A test built for the wrong overlap type returns clean. Sitting through 10 minutes of steady-state on a start-overlap fault proves nothing, and vice versa.

Step 2: Find the shared node, and plan to measure there

The equipment is almost never where the fault lives. The fault lives at the thing the two actors share, and that is where the meter, gauge or probe goes.

Common shared nodes by trade: a branch circuit, a feeder, a neutral or a service; a supply pipe, a pressure tank, or a water heater's recovery capacity; a common vent or a combustion air opening; a shared drain or condensate line; a control transformer, a communication bus, or a common low-voltage supply. Identify yours before you touch anything, and confirm it rather than assuming it, because the boundary of a shared circuit or a shared line is exactly the thing sites get wrong.

Step 3: Baseline each actor alone, at the shared node

Three readings at the same point, in this order: nothing running, actor A alone, actor B alone. Same probe position, same meter, same units, all three. Move the probe between readings and the whole comparison is worthless, because you will be comparing a difference in load against a difference in test point.

The no-load reading is the reference everything else is measured against, and it is the one people skip. Without it you have two numbers and no way to say how far either one moved.

Step 4: Set the pass and fail rule before you run the pair

Decide the threshold while you still have no result, so the result cannot bend it. For an electrical shared node, a workable rule, measured at the equipment terminals and referenced to the no-load reading taken at that same point:

Flag the shared node when the combined-load deviation exceeds 5% of the no-load reference AND exceeds 1.3 times the sum of the two individual deviations. The 5% half comes from the NEC informational note; the 1.3 is a working default you set and tune against your own confirmed finds, not a published figure. The Boolean is AND, and both halves matter for different reasons. The 5% half comes from the conductor-sizing guidance in NFPA 70, Article 210, whose informational note recommends limiting branch-circuit voltage drop to 3% and total feeder-plus-branch drop to 5%; informational notes in that document are advisory, not enforceable requirements, so treat 5% as an engineering flag rather than a code violation. The 1.3 half is what separates a defect from a capacity limit, and it is the more diagnostic of the two.

Why non-additivity is the real signal: if a run is simply loaded near its design limit, the drop under both loads lands close to the sum of the individual drops. A connection with abnormal resistance behaves worse than additive, because the extra current both raises the drop across that resistance and heats it, which raises the resistance further. Additive means sizing. Worse than additive means a defect, usually a specific joint.

For a hydraulic shared node the same shape applies to pressure or flow, and for a thermal one to temperature rise. The units change, the rule does not.

Step 5: Run the pair, and hold it

Bring both actors up with the overlap type you named in Step 1, and measure at the shared node throughout. For start overlap you need a meter that captures a short event, so use a min-max or peak capture function rather than watching a display. For steady-state overlap you need dwell: run both for several minutes and take the reading at the end of the dwell, not at the start, because a resistive or thermal fault gets worse over the first few minutes and a reading taken at second thirty will understate it.

Step 6: Confirm the direction by removing one actor

Drop actor B while the meter is still on the shared node and watch the reading recover. A fault that resolves the moment one actor leaves is confirmed as concurrency-dependent. A reading that stays bad after B drops means you have found a second, unrelated problem, and you now have two findings rather than one.

This step is short and it is the one that survives a challenge. Anyone can produce a bad reading. Producing it, removing one load, watching it clear, and reproducing it again is a demonstration.

Worked example: two loads that are each fine

The complaint is that an appliance runs badly and sometimes quits, and that it seems to happen in the evening. Single-load testing by a previous visit found nothing. The suspected shared node is a branch circuit serving both that appliance and a second sizeable load.

All readings taken at the appliance's terminals, same probe position throughout, on a nominally 120-volt supply.

• No load, nothing running: 121.0 volts. This is the reference.
• Appliance alone: 118.1 volts. Deviation 2.9 volts, which is 2.4% of the 121.0 reference.
• Second load alone: 117.6 volts. Deviation 3.4 volts, which is 2.8% of the same reference.
• Both together, after a 5-minute dwell: 110.5 volts. Deviation 10.5 volts, which is 8.7% of the reference.

Now run the stated rule. The combined deviation is 8.7%, which exceeds 5%. The two individual deviations sum to 6.3 volts, and the observed combined deviation of 10.5 volts is about 1.7 times that sum, which exceeds 1.3. Both halves of the AND are satisfied, so the shared node is flagged as a defect rather than a capacity limit.

Note what each actor looks like alone: 2.4% and 2.8%, both under the 3% branch-circuit recommendation, both entirely unremarkable. A tech who tested either one on its own would have written "voltage good at the equipment" and been telling the truth. That is precisely the blindness this method exists to close.

Step 6 confirms it: dropping the second load returns the appliance terminals to 118.0 volts within a couple of seconds, essentially back to its solo reading. Re-adding it takes it back down. The behavior is repeatable and it tracks the pair, not either actor.

The non-additive result points the search at a joint carrying both loads, so the search runs from the appliance back toward the source, and the section that shows the largest drop per unit of length under load contains the defect. With the circuit de-energized, locked, tagged and proven dead, terminations along that section get inspected and remade.

Change one number and the conclusion changes. Had the combined reading been 114.6 volts, the deviation would be 6.4 volts, or 5.3% of the reference, which exceeds the 5% threshold but sits at only about 1.02 times the 6.3-volt sum. It satisfies the percentage half and fails the ratio half, so under the stated AND it is not flagged as a defect. It is a correctly-made circuit carrying more than it was sized for, and the honest recommendation is load redistribution or a dedicated circuit, not a hunt for a bad joint. Same complaint, same test, different repair, and the thing that separates them is the ratio, not the percentage.

The failure mode worth naming: a tech who takes the combined reading without the no-load reference has 110.5 volts and nothing to compare it to. Called against nominal 120, that is a 7.9% deviation, which still trips the 5% flag but gives no access to the ratio test at all, so the defect and the overload case become indistinguishable. The reference reading takes about 20 seconds and it is the only thing that makes the rest of the numbers mean anything.

How to verify the test was valid

• Same point, same meter, all four readings. If the probe moved between the baselines and the combined reading, discard the set and redo it. This is the most common way a good method produces a confident wrong answer.
• The overlap type you tested matches the one you named. A steady-state dwell run against a start-overlap complaint returns clean and proves nothing. If the customer's description says "when it kicks on," you need capture, not dwell.
• You demonstrated it twice in both directions. Bad with both, good with one, bad with both again. One occurrence is an event.
• Your write-up shows the reference reading and the ratio, not just the bad number. The next tech, the warranty reviewer and the electrician who takes the repair all need to see how you separated a defect from an overload, because those two findings lead to completely different work.

References

• NFPA 70, Article 210, branch-circuit conductor sizing and its informational note recommending 3% branch and 5% total voltage drop; informational notes in that document are advisory rather than enforceable
• 29 CFR 1910.333(b)(2), OSHA electrical safe work practices, for de-energizing and locking or tagging before work on electric circuit parts
• NFPA 70E-2021, 120.5, the live-dead-live process for establishing an electrically safe work condition, and its energized-work provisions where a measurement cannot be taken de-energized
• 29 CFR 1910.147, OSHA control of hazardous energy, for isolating stored energy before mechanical service
• See related: How to Map Which Loads Share a Circuit or Supply; The Fault That Only Exists Under Concurrent Demand