You Can't Reproduce What the Customer Does: Decision Tree

Why this matters

The customer swears it fails, and it works perfectly the whole time you are there. Before you write "no fault found" and leave, understand what that usually means: you are not reproducing the conditions the fault needs. Real operating life is not a five-minute test in good light with the system half-loaded. The fault is often real and waiting for a condition you have not recreated: a load, a duration, a temperature, a sequence. This tree is how to reconstruct the customer's actual use so the fault has a reason to show itself.

Start here: if the report was a hazard, treat it as real

A symptom you cannot reproduce is not automatically a non-event.

If the customer reported a hazard - a shock, a burning smell, a spark, a gas odor, a scald, an overpressure - do not dismiss it because it behaved for you. An intermittent hazard is still a hazard. Inspect for the evidence it leaves behind (heat discoloration, arcing marks, moisture, corrosion, a tripped safety), and where you find a credible hazard mechanism, make it safe and address it even though it did not perform on cue. Absence of a repeat in ten minutes is not proof of safety.

If the symptom is not a hazard, work the reproduction problem below.

First, reproduce their exact conditions, not a convenient version

The usual reason it "works for you" is that you tested it your way, not their way. Close the gap:

• Ask them to operate it, in their normal conditions, while you observe. (This overlaps with an operator-specific fault but is not the same thing; here you are recreating conditions, not just watching a person.)
• Match the real load. A system tested empty or half-loaded hides faults that only appear under full demand.
• Match the real duration. If it fails after an hour of running, a five-minute test proves nothing. Let it run.

Walk the four condition variables

Faults that hide from a short test are almost always gated by one of these:

• Load - it only fails when fully loaded, or when a second thing runs at the same time.
• Time / duration - it needs to heat up, cool down, fill, or run long enough to fault.
• Temperature / environment - it only fails in the cold morning, the hot afternoon, or high humidity.
• Sequence / cycle - it only fails on a specific start, stop, or combination the customer does and you did not.

Change one at a time and hold it long enough. Reproducing the fault once, deliberately, is worth more than any amount of "it seems fine."

If you still can't reproduce it

When the conditions resist recreation on this visit:

• Look for the fault's fingerprint. Even a fault that will not perform leaves traces: heat marks, wear, residue, a fault memory, a witness mark. Read the aftermath.
• Instrument it and leave it. Where the trade allows, set up monitoring, or ask the customer to capture it (note the time, the conditions, a photo or short video) so the next visit starts with data instead of a mystery. Tell them exactly what to record.
• Do not swap parts to feel productive. Replacing components against an unreproduced fault is the parts-cannon: expensive, unverifiable, and it destroys your credibility when the symptom returns.

Decide the honest close

Two legitimate outcomes:

• Found and fixed - you reproduced it, diagnosed it, and verified the repair against the same condition that triggered it.
• Not reproduced, monitored - you could not make it fail, you documented the conditions to capture next time, and you told the customer plainly where things stand. That is an honest result; a confident guess is not.

Recap the order

1. Safety first. If the report was a hazard, treat it as real and inspect for evidence even if it will not repeat.
2. Reproduce their exact conditions, not a convenient short test.
3. Walk load, time, temperature, and sequence, one variable at a time, held long enough.
4. If it still hides, read the fault's fingerprint and instrument it; do not swap parts blindly.
5. Close honestly: found-and-fixed, or not-reproduced-and-monitored.

The machine is not lying to you, and neither, usually, is the customer. You just have not yet asked it to fail the way real life does.

References

• Trade-standard intermittent-fault practice: reproduce under real load and duration before concluding no fault found
• Manufacturer documentation on operating conditions and fault-memory data where available
• See related: The Fault Follows the Operator, Not the Equipment (decision tree); Separating a Usage Habit From an Equipment Fault