Reduce a failure example without claiming its cause

A large example can contain many details unrelated to the failure being investigated. Reducing it helps another person inspect the behavior, but a small example does not explain its cause. Use the stipulated fictional observations below to practice a deletion trace. They are invented results for a reasoning exercise, not runs of Kiguri or another product.

Define the observation and permitted deletions

Label six self-contained document fragments A through F. For this toy, any retained subset remains a valid input, and retained fragments keep their original relative order. Define fail as the same fictional output discrepancy occurring; pass means that discrepancy is absent. Keep an unknown category for a result that cannot be assessed.

Record the untouched input ABCDEF as fail. Preserve its contents and the observation definition beside the reduction trace. Do not redefine the failure as any error, because a smaller invalid document might produce a different problem. In real investigation, validity and repeatability need evidence; here they are explicit exercise assumptions.

Compare subsets without losing interactions

The supplied observations say ABCD passes and EF passes. That does not contradict ABCDEF failing. A detail in one subset may matter only when another detail is present. Do not conclude that neither half contains useful information or that the original observation must be mistaken.

The next stipulated observation is that ACDE fails. Retain that smaller failing input and note that B and F were removed together. This establishes only that their absence did not remove this toy discrepancy. It does not establish that they are irrelevant to every failure or that either remaining fragment is the cause.

Continue until the remaining deletions stop reproducing

The exercise then supplies CE as fail, while C alone and E alone both pass. Write those three observations explicitly. CE is a failing input for which each permitted single-fragment removal loses the defined discrepancy. Under these tested deletions, it is minimal against single-fragment removal.

That result is not a search through every possible alternative input. Another different example could be smaller, or another fragment combination could reproduce the discrepancy. Do not invent a hidden formula saying C and E cause the problem. The trace supports a reduced example and its tested deletion property, not a unique sufficient cause or an explanation of implementation behavior.

Keep a reproducible reduction record

For each row, record the parent input, retained identities, removed identities, validity assumption, result, and reason for the next selection. A reader should be able to reconstruct ABCDEF to ACDE to CE without guessing which observations were used. Keep the passing complementary subsets too; they explain why simply choosing a failing half was not available.

In an actual investigation, repeat the final observation separately before handing it over. If results change between repetitions, mark the example unstable and investigate that uncertainty rather than declaring minimality. Do not replace an unknown or invalid result with pass merely to finish the trace.

Check the handoff's claim

Ask another person to explain what CE establishes. They should identify the stipulated fail, the two passing single deletions, and the limits of the search. If they describe CE as the only root cause, correct the summary.

The useful artifact is an inspectable reduced reproducer with its validity and observations attached. It makes a next investigation smaller while leaving diagnosis open. A compact example is helpful precisely when its reduced scope does not turn into a stronger claim than the evidence permits.