Use accepted and rejected examples together.
Infer a candidate criterion using positive and negative rows. Review exact counts, positional roles and counterexamples without personality or ability labels.
Six original items each contribute one correct answer. A count out of six describes this fixed bank, without an ability band, percentile or employer threshold. All six explanations and this full guide remain available at every count.
One property must explain both sides
A classification question marks some rows IN and others OUT. A useful rule must accept every positive and reject every negative, within the candidate criteria the prompt explicitly supplies. A shared feature of the positive rows alone is not enough. Several properties can coexist, such as containing a repeated token and being a palindrome. The negative rows show which broader properties are insufficient. Keep the labels as evidence rather than assuming that an attractive visual pattern must be intended.
A fresh counterexample separates two rules
Suppose PQQP is IN but PQRP is OUT. Both have equal outside tokens, so matching ends cannot explain the labels. Full palindrome can: reversing PQQP preserves it, whereas reversing PQRP gives PRQP. This does not establish that every possible classification uses palindromes; it only compares the supplied candidates against these facts. For a new target, inspect the complete row rather than checking the first and final tokens and stopping early.
Exactly one differs from at least one
Quantifier words matter. A row with three As satisfies at least one A but fails exactly one A. If IN rows each contain a single A while an OUT row contains two, the latter is evidence against the broader criterion. Count positions, including repeated occurrences, rather than counting only the distinct symbol kinds. The question might instead ask for two different tokens, which is a separate property; do not exchange those definitions because both involve counting.
A repeated token may occupy the wrong role
Under a leading-pair rule, XXYZ has its equal tokens at the start and two different remaining tokens. XYXZ contains the same number of Xs but puts them in different positions. A candidate criterion may also require Y and Z to differ from X and from one another. Check every clause: a row such as XXYY has a leading pair but fails that final distinctness requirement. The specific role constraints distinguish the correct target from a superficially similar one.
Build a small rule-by-example table
Give each candidate rule a row and each labeled example a column. Mark whether the predicted label agrees with the given one. Cross out a rule as soon as one disagreement occurs, then verify the survivor against the remaining examples. Only after that should you evaluate the answer options. This avoids fitting a rule to whichever target you already prefer. You can use a paper table; no statistical learning model or hidden classification service runs in this practice.
Construct a near-miss target
After reading the key, create one new row that satisfies the criterion and one that misses just one requirement. For instance, if two inside positions must differ, make them equal while keeping the outside positions unchanged. Explain exactly which condition fails. A near miss is more useful than a completely unrelated string because it isolates the distinction you need to learn. Retaking the same six known items can reward recall, so keep new examples separate from the reported attempt.
Classification criteria are local definitions
The six counts reflect how you used the explicit symbolic criteria in this edition. They are not personality categories, diagnostic findings, general-intelligence estimates or evidence of cultural neutrality. A finite collection of labeled examples cannot establish a unique rule among all imaginable functions. Our item checks establish one surviving criterion within each stated finite candidate list and one correct answer option. If a real classification task lacks that boundary, acknowledge ambiguity and seek more evidence rather than claiming certainty from a few examples.