How to Narrow a Fault to a Branch Without Guessing

Why this matters

Halving a system to find a fault is the oldest method there is and it works, right up to the moment somebody applies it to a distribution circuit. Then it quietly stops working, because on a distribution circuit the act of splitting changes the quantity the protective device is measuring. Turn off half a panel to see if the feeder still trips and you have not isolated anything; you have removed half the load from the device that was reporting the load. The circuit holds, everyone relaxes, and the answer you write on the ticket is wrong.

The sibling article on divide and conquer covers the general logic of halving a chain. This one covers the two conditions that make it valid on a supply: the topology has to come from evidence rather than from a label, and the load state has to be held while you split.

Before anything: the work-practice gate

Most of the readings below exist only with the system energized. 29 CFR 1910.333(a)(1) requires de-energizing before work on or near exposed energized parts unless the employer can demonstrate that de-energizing introduces additional or increased hazards or is infeasible by equipment design or operational limitation, and that demonstration lives in the employer's program rather than in a decision made at the panel. Where live work is justified, the approach boundary and arc-rated PPE come from NFPA 70E-2021 130.5 and 130.7 in the edition your employer's program has adopted. For isolation, panel and branch-circuit lockout runs under 29 CFR 1910.333(b)(2) in general industry or 29 CFR 1926.417 on construction work, since 29 CFR 1910.147 excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations at (a)(1)(ii)(C), and proving dead follows NFPA 70E-2021 120.5.

And name the circuits you may not switch for a test. Do not de-energize or deliberately re-fault a circuit serving fire alarm, emergency lighting, egress, life-safety equipment, or a process whose loss of power creates a hazard, without the owner's written coordination and whatever fire-watch or standby the facility requires. Do not deliberately re-create a condition on a circuit with a suspected arcing fault or a confirmed ground fault; those get repaired first and tested after.

Step 1: Get the topology from evidence, not from the directory

Panel schedules are historical documents. Renovations tie circuits together, spare breakers get used, and labels stop matching within a few years of occupancy.

With each device off in turn and the circuit proved dead, confirm what is actually dead. Write what you find on the schedule as you go. This step feels like a delay and it is the only reason the rest of the procedure produces a true answer, because everything downstream depends on knowing which conductors belong to which device.

Skip it and you get: a correct measurement attributed to the wrong branch, which is worse than no measurement, because it survives into the next tech's assumptions.

Step 2: Define the fault-present state and write it down

Before any splitting, record the exact conditions under which the fault happens: which loads are running, the total current, the enclosure ambient, and the elapsed time from energization to the event. The time band tells you which element operated and therefore what the device is even measuring, and the article on reading a trip as evidence covers how to read those bands.

Skip it and you get: no way to tell later whether your test recreated the fault condition or merely resembled it.

Step 3: Ask whether your split changes the measured quantity

This is the step that does not exist in ordinary divide and conquer, and it is the one that matters.

• A thermal overcurrent trip is a function of total circuit current. Any split that removes load changes the input. The test is invalid before you run it.
• A magnetic trip is a function of a fault that either is or is not on the half you kept, so splitting works cleanly here.
• A ground-fault trip is a function of summed leakage from everything downstream, so splitting changes the input again, in the same way and for the same reason.
• A thermal condition at the device rather than in the conductor is a function of enclosure heat, and switching half a panel off changes that too.

If the quantity is a total, a split that reduces the total answers a question nobody asked.

Skip it and you get: the false clear demonstrated below, where both halves pass.

Step 4: Preserve the quantity, or measure something independent of it

Two ways out, and one of them is always available.

Preserve it. Re-sequence which equipment runs so the total stays at the recorded value while the set of branches supplying it changes, or substitute an equivalent temporary load, which is its own energized connection and belongs in the same authorized window under the same gate, landed on a de-energized and proved-dead circuit before anything is re-closed. Then the total is constant and the only variable is which branch is in circuit, which is what you wanted to test in the first place.

Or measure something that does not depend on total load. A per-branch differential leakage reading, a de-energized insulation resistance test, and a comparative thermal reading are all independent of how much of the panel is switched on. For thermal comparison, the rule is to compare identical components at similar current against each other rather than against a remembered absolute number: a lug running noticeably hotter than the identical lug beside it at similar current is the signal, and any absolute limit belongs to the equipment manufacturer or the adopted standard.

Skip it and you get: an isolation exercise that can only ever produce ambiguous results on the most common trip type there is.

Step 5: Halve at documented junctions, not at the middle of a run

The useful split points on a radial system are the places where the circuit fans out - a panel, a junction, a disconnect, a terminal block - because those are the points you can restore exactly and describe to the next person. Cutting into the middle of a homerun to halve the wire length is a repair you now own, and it adds a splice that becomes a suspect on the next call.

Skip it and you get: a system with more joints in it than it started with and a record nobody can reconstruct.

Step 6: Prove the branch by reversal, both directions

A fault that disappears when you remove a half has told you very little. A fault that follows the suspect half when you move it to another supply, and stays away from the original when you restore everything else, has told you where it lives.

Run it both ways or do not claim it: suspect half out, condition restored, no fault; suspect half back, condition restored, fault returns.

Skip it and you get: a branch convicted on absence of evidence, which is how the same circuit gets diagnosed three different ways by three techs.

Step 7: Localize inside the branch by the same rule

Once the branch is proved, the same two conditions apply at finer scale. Split at the accessible junctions inside it, hold the branch's own load state, and prove by reversal again. Nothing about the logic changes; only the scale does.

Skip it and you get: a whole branch rewired to fix one bad connection.

Worked example: the halving test where both halves passed

A 100 A feeder to a subpanel, tripping after about two hours of normal operation. Two visits already, no fault found.

Step 2 first. Under the normal operating condition, the total feeder current measures 86.0 A. Against the 100 A device that is 86 percent of rating. Read that as a sizing finding rather than a trip explanation: in the NEC edition your authority having jurisdiction has adopted, the device must be rated at not less than the noncontinuous load plus 125 percent of the continuous, so 86.0 A continuous needs at least 107.5 A. Equivalently a standard device takes no more than 80.0 A of continuous load unless the assembly is listed for 100 percent operation. This feeder sits 6.0 A past that, which is real and is not why it opened.

The halving test the previous tech ran. Six branch circuits, measured individually at the same moment as that 86.0 A total: 22.0, 18.5, 14.0, 12.5, 11.0 and 8.0 A. They sum to 86.0 A, confirming the readings are consistent and nothing unmetered is on the feeder.

• Switch off branches 4, 5 and 6, leaving 54.5 A, which is 25.5 A below the continuous limit. The feeder holds all day. Conclusion drawn: the fault is in branches 4, 5 or 6.
• Switch off branches 1, 2 and 3 instead, leaving 31.5 A. The feeder holds all day again.

Both halves passed. Impossible if the test were valid, and the clearest demonstration of step 3: the split was changing the exact quantity the thermal element measures. Whatever made this device open at 86 percent of rating, nothing makes a correctly built thermal element open at 54.5 or 31.5 percent, fault or no fault. State both ends of the range and the test collapses on inspection.

What the device was reporting, and what it was not. Not a fault, and not a simple overload. Separate the two findings. The sizing finding stands on its own: the device is undersized for this load whether or not it ever opens. But a 100 A device is calibrated to carry 100 A, so 86.0 A is not an overload and a healthy inverse-time element does not operate there in two hours or in twenty. The 80 percent figure exists BECAUSE a device in an enclosure runs hotter than its open-air calibration assumed, which points at the field recorded in step 2 and never used. Read the enclosure ambient and the lug and bus temperatures against identical neighbours, and confirm the device's condition: a trip at 86 percent of rating is the enclosure and the terminations reporting, on top of a load the panel should not carry. Both go on the ticket. Only one is answered by moving branch 4.

Which sends the work back to step 1. Verifying what each breaker actually kills, rather than what the schedule claims, finds branch 4, labelled a small lighting circuit, feeding receptacles in an adjacent area that a renovation tied into this panel. Its 12.5 A does not belong to this feeder.

And the arithmetic closes. Remove branch 4's 12.5 A and the feeder carries 73.5 A, 6.5 A under the 80.0 A continuous limit. Returning that load to the panel it was designed for is the repair, and it is one somebody has to design and permit rather than a reconnection to make on the spot.

The failure mode of the original approach: with the false clear pointing at branches 4, 5 and 6, the next step would have been an insulation test and a teardown of three healthy circuits, finding nothing, followed by a recommendation to replace the feeder breaker. A new device might well have behaved the same way, because nobody would have read the enclosure or the terminations that made this one open below its rating.

The condition that flips this. A trip at energization rather than two hours puts you in the magnetic band, where a split genuinely does isolate, because the fault either is or is not on the half you kept. Same panel, same procedure, opposite validity, and the deciding fact is the timing from step 2.

How to verify you got this right

• Check that your test could have failed. Both configurations passing, or both failing, means the split was not measuring location. Say so on the ticket rather than picking the half that fits.
• Confirm the total in every configuration, not just the first. Writing 86.0, 54.5 and 31.5 next to each other is what makes the false clear visible. A tech who records only the pass or fail never sees it.
• Restore and re-fault once, deliberately, before you leave - on a circuit that the exclusions in the work-practice gate above permit: no fire alarm, emergency lighting, egress or life-safety load, no suspected arcing fault, no confirmed ground fault, and with the owner's written coordination for the window. If your identified condition is right, the fault returns on schedule, which is the strongest confirmation short of the repair itself.
• Reconcile the branch measurements against the total. If the branches do not sum to the feeder reading, something is connected that you have not accounted for, and that load is now the most interesting thing on the job.

References

• 29 CFR 1910.333(a)(1) and (b)(2); 29 CFR 1910.147(a)(1)(ii)(C); 29 CFR 1926.417 for construction work
• NFPA 70E-2021, 120.5, 130.5 and 130.7, as adopted by your employer's electrical safety program
• NEC Articles 210, 215 and 240, including the continuous-load provisions, in the edition your authority having jurisdiction has adopted
• See related: Divide and Conquer: Isolating a Fault; How to Read a Trip as Evidence; Why Nuisance Tripping Is Usually a Real Signal