Penalty.kind

Penalty.unitEnd is the exclusive end of a measured unit span. It must be greater than zero and no greater than the paragraph unit count. The penalty's position in the Item array does not set that boundary.

Planning collects penalties and visits them in increasing unitEnd order. Equal endpoints retain input order. Duplicate endpoints remain separate candidates and are counted separately in breakOpportunityCount.

Penalty.rule determines whether a candidate is selectable and how its additional cost is interpreted. Forbidden removes that candidate. A selectable duplicate at the same endpoint remains available. A Forced entry prevents any line from crossing that boundary.

Penalty.kind describes the selected boundary independently of its rule. For example a Forced rule can carry a LatinHyphen kind. The chosen line reports the selected candidate's kind.

Penalty.width is a nonnegative Length<Pt> that contributes when that candidate ends a line. Both cost models include this width for every kind. Unselected candidates contribute no width.

PlanLine.selectedPenaltyWidth reports that selected width separately from the measured text widths. Input penalty widths are validated even for Forbidden candidates.

Penalty.flagged marks a break for KnuthPlassPolicy.flaggedDemerit and finalFlaggedDemerit. FixedPoint does not use this flag. Its consecutive soft hyphen rule depends on kind.

An out-of-range unitEnd or negative penalty width returns InvalidInput from breakParagraph. Record and Item construction preserve the supplied data until that call validates it.

A complete plan needs a selectable terminal candidate at the unit count. Paragraph planning does not synthesize a terminal penalty or invoke break classification on the measured items.