How to Diagnose a Long Supply Run Rather Than the Equipment

Why this matters

The detached shop at the back of the property, the fixture at the far end of the addition, the unit on the far side of a long crawl, the appliance the customer moved to the other end of the building. In every trade these addresses generate the same story: equipment that tests fine, gets replaced, and fails again. The equipment was never the problem. It is starving at the moment it asks for the most, and everything between it and the source is where the loss happens.

The reason this fault survives so many visits is that it is invisible at rest. At zero demand a long run measures almost perfect. It only shows itself under load, at the far end, and almost nobody measures at the far end under load.

Before you measure anything

Lead with the isolation that matches the medium. You are about to work at both ends of a run, often in a crawl, an attic, or a detached structure, which means you may be out of sight of the other end while somebody else is near it.

• Electrical. Open the disconnect, lock and tag it, and prove dead with a meter checked against a known live source immediately before and immediately after, which is the live-dead-live sequence in NFPA 70E-2021, 120.5. The duty to de-energize and lock or tag before working on a circuit that could become energized is at 29 CFR 1910.333(b)(2) for general industry, and at 29 CFR 1926.417 where the same electrical work falls under construction rules. If part of the work genuinely has to be done energized, that is a decision with its own justification, its own protective equipment, and a second person aware of it, not a convenience.
• Pressurized systems and stored energy. Isolate, block and relieve stored energy before opening a run, per 29 CFR 1910.147, and confirm relief on a gauge rather than by the position of a valve handle. A long run holds a lot of stored volume and it does not empty because you closed something at one end.
• Gas. Gas odor means everyone leaves the building immediately, no switch is touched, no light is turned on, no phone is used inside, and the call goes in from outside. Any joint you open on a gas run gets leak-tested with a listed leak-detection solution or a combustible-gas detector before anything is fired.
• Two ends, two people. Tag both ends. A lockout at the source is what stops the person at the far end from restoring what you isolated.

Step 1: Find the demand at which it actually fails

Do not start measuring. Establish the condition first: ask what else is running when it happens, and what the equipment is doing at that moment, whether that is starting, at peak output, filling, firing, or at the end of a long cycle. Skip this and every reading afterward is taken at the wrong condition, which is exactly why the last two techs passed it.

Step 2: Name your two points

The source is the last place the supply is common to everything else: the panel, the meter, the manifold, the main. The point of use is the equipment's own connection, past every last fitting, splice, valve and terminal. Not the receptacle nearby, not the shutoff under the unit, not a convenient test port halfway. Anything short of the equipment's own connection leaves an unknown length of run inside your "equipment" and outside your "run," and that unknown is often the whole fault.

Step 3: Measure both points at the same instant, under that demand

This is the entire method, and it is where it gets skipped, because it needs two instruments or two people. Two readings taken sequentially are not a comparison: the source itself moves as other loads come and go, and a long run's loss can change by a large fraction in the time it takes you to walk back. If you have one instrument, leave a logger at one end or put a helper on a phone at the other and count down to a shared instant.

Take three sets: at rest, at normal demand, and at the demand where the fault appears.

Step 4: Turn the pair into two comparisons, not one

Compute the loss across the run as a percentage of the source reading at the same instant. Then compare that loss against two separate numbers, because they answer different questions:

• Against the design allowance for a run of that type, which tells you whether the run is built correctly.
• Against the equipment's own stated tolerance at the point of use, which tells you whether the equipment can live with it.

Those can disagree, and the disagreement is informative. A run inside its design allowance feeding equipment with an unusually tight tolerance is an equipment-selection problem. A run well outside its allowance feeding forgiving equipment is a fault waiting for the next appliance the customer adds.

On the electrical side, the widely used design target is about 3% loss on the branch circuit and about 5% combined across feeder plus branch, which appears in the National Electrical Code as an informational-note recommendation rather than an enforceable requirement, so treat it as the number good practice aims at rather than something an inspector fails you on.

Step 5: Classify the run from how the loss behaves

Three shapes, three different repairs. Distinguish them by how loss changes with demand and by where it sits along the run.

The first row is the one people miss, because it is not a failure. Nothing broke. The run was always at its limit and the customer added a load, changed a schedule, or replaced equipment with something that draws harder at startup.

Step 6: Localize, only if the shape says to

If the loss is proportional and distributed, there is nothing to localize and segment testing is wasted time. If it is degraded or stepped, split the run at its midpoint, measure there under the same demand, and split the offending half again. Choose midpoints where access already exists rather than where the geometry is tidy.

Step 7: Pick the remedy against the two comparisons

Four options, and more than one is usually valid:

• Reduce the concurrent demand on the run. Rescheduling a load off the peak window costs nothing and is the fastest way to confirm the diagnosis before anyone commits to work.
• Repair or upsize the run. Correct for a degraded or stepped shape, necessary for a proportional shape when demand is not negotiable.
• Move the equipment or the supply point closer. Sometimes the cheapest real fix, especially when the far end is a recent addition.
• Support the equipment locally with the trade-appropriate device that restores the missing quantity at the point of use, where that is an accepted practice for the system in question.

What is not on the list: replacing the equipment. It cannot help, and knowing why is the point of step 4. The replacement carries the same tolerance as the original and will meet the same shortfall.

Worked example: the third unit at the back of the property

Detached structure at the far end of a long run. Two units had already been replaced at this address, both of which reproduced the fault within a week. The complaint is that the equipment drops out when it works hardest.

Readings, taken at both ends at the same instant, expressed as a percentage of nominal:

Two things come straight out of that table.

The equipment is being starved, and it is not marginal about it. Its data plate states a minimum of 90% of nominal at its own connection. Under the condition where the customer's complaint occurs, it is receiving 88%, which is 2 points below what it is built to accept. It is doing exactly what it is designed to do when underfed.

The run is the cause and it is undersized rather than broken. The loss is 1 point at rest, 4 points at normal demand, and 9 points at fault demand, rising with demand rather than sitting at a fixed value. As a share of the source reading at that instant, 9 points against 97 is a 9.3% loss, which is about 3 times the roughly 3% branch design target. Nothing about that pattern says a bad joint. A single degraded connection produces a step at one place, and segment readings along this run showed the loss distributed evenly.

Why replacing the equipment failed twice. Both replacements carried the same 90% minimum, because that is a normal tolerance for that class of equipment. Handed 88%, they behaved identically. Two replacements, two identical outcomes, and the shop's own record was the strongest evidence in the file that the equipment was not the variable.

How they proved it before spending anything. They removed the concurrent load that was pushing demand to the fault point and re-measured at the equipment under the reduced demand. It came in at 94%, above the 90% minimum, and the fault did not occur across the following week. That single reversible change confirmed the diagnosis without touching a wire.

What was actually done. The run was corrected so it carries the load it is asked to carry. Post-repair, at the same fault-condition demand, the equipment read 96% against its 90% minimum, which is 6 points of margin.

Why 6 points and not "it passes." The source itself moved between readings, from 100% at rest to 97% under load, and that is normal, seasonal, and outside your control. A repair that lands the equipment 1 point above its minimum has no room for a hot afternoon or the next load the customer adds. Set a margin rule and hold to it: at the worst demand you can construct, require the point-of-use reading to sit at least 5 percentage points above the equipment's stated minimum, measured at the equipment's own connection.

The same shape in other trades. Water: near-static pressure at the far fixture with nothing running, and a large drop the moment flow starts. Gas: correct pressure at rest, low at the firing rate. Air distribution: acceptable at low output, starved at full. In each, the loss is a function of demand, which is precisely why every test conducted at rest passes and why the equipment gets blamed.

How to verify you got this right

• Both readings were simultaneous, and taken at the equipment's own connection rather than a convenient point upstream. A sequential pair is the most common way this method produces a wrong number that looks right.
• You have three conditions, not one. Rest, normal, and fault-condition. The shape across those three is what classifies the run, and a single reading cannot classify anything.
• You compared against both numbers, the design allowance and the equipment's stated tolerance, and wrote both into the record beside the readings, with the margin stated at the worst demand you could construct.
• You reversed something to confirm before you committed. Dropping the concurrent demand and watching the reading recover changes exactly one variable, and it is worth more than any amount of confident arithmetic.

References

• National Electrical Code informational-note recommendations on voltage drop, roughly 3% on a branch circuit and roughly 5% combined for feeder plus branch, which are recommendations rather than enforceable requirements
• 29 CFR 1910.333(b)(2), and 29 CFR 1926.417 for the same electrical work under construction rules, on de-energizing and locking or tagging before work
• NFPA 70E-2021, 120.5 for the live-dead-live instrument proving sequence
• 29 CFR 1910.147 on isolating and relieving stored energy before opening a pressurized run
• See related: The Half-Split Method on a Long Run