F.18:4.3 - Candidate Selection

Do not pick a durable label in one stroke or work toward a fixed candidate count. Unless a cited external standard fixes the label under §4.1, build the smallest set that covers at least two live head-term families. In either branch, examine every plausible neighbouring-object reading that could change the decision. Stop when each live family has a representative and no untested plausible alternative could overturn the selection. If a deadline forces closure while a plausible family or alternative remains untested, record that exception in CandidateCoverage and make it part of RefreshCondition.

Judge candidates on:

• semantic fidelity: does the label preserve the governed value without adding or losing required conditions?
• reader ergonomics: can the intended reader recognize, say, and remember it in the current situation?
• morphology fit: does the word shape fit the kind being named, for example an exact local system-role kind, method, work, description, relation, slot, characteristic, or status value?
• alias risk: will a careful reader import a wrong sense from nearby FPF patterns or external practice?

Use these as ordinal comparisons. Do not average them into one score. If a Pareto-front or quality-diversity method is used, the dimensions and dominance rule must be visible on the card.

One candidate can win even when it is not perfect, but the SelectionRationale must say what it buys, what risk remains, and why the covered set is sufficient for this use.