Precedent: boundaries

What the construct does NOT do: limits, reserved values, neighbouring constructs and the selection rule.

OverviewBoundariesCorpus formsPitfalls

Does not do

• Not know the word argue. There is no argue keyword in the grammar (verified against the grammar during this research). Side argumentation as a construct is absent: the “case differs” point is expressed by the precedent’s factor sets and case facts, not by a separate declaration.
• Not know courts as entities. There is no standalone court declaration; court is only an item inside precedent, naming a preorder vertex from the vocabulary’s courts. A court’s characteristics (bench, jurisdiction, instance outside the line) are invisible to the profile.
• Not executed by ordinary queries. The profile’s own answer (the BINDING/DISTINGUISHED/INAPPLICABLE/JUDGMENT_PENDING/NOT_YET/ OVERRULED statuses, constraint, caseBase) is available ONLY via the precedent query kind. truth queries see only baseline execution of the generated holdings — no status names, no distinguishedBy/missing.
• Not weigh or measure similarity. Binding is set inclusion; factor magnitude and similarity metrics are outside the v1 slice. Two cases are either linked a fortiori or named as differing — no halftones.
• Not repeal another holding. Repeal is declared by the repealed precedent in its own package as an overruled (date, reference) pair; there is no cross-package overrules. Lifecycle ownership is as with editions.
• Not recompute agreesWithRules in the core. The v1 checking core verifies statuses, constraint and caseBase, but not the rule-declaration agreement flag — the named boundary (needs baseline computation inside the core).
• Not check maximality of preferred extensions — but that is the neighbouring argumentation profile’s boundary, not precedent’s.
• Not cut the past off silently. A holding holds from decided inclusive; the date is mandatory exactly because without it the holding would have no force window. Edition projection lifts the holding like a lapsed edition.

Reserved and closed

• The vocabulary is closed: a factor undeclared in it does not exist for the profile. Violation — LDC-E2701, a factor on both sides — LDC-E2702.
• The outcome is a literal with polarity over signature variables (LDC-E2703); ratio is a subset of the winning side’s X_s (LDC-E2704).
• limits.maxPrecedents defaults to 64; overflow is fatal RESOURCE_LIMIT. No partial conclusion is published then.
• Generated-node identifiers are deterministic: <P>/holding, <P>/over/<R>, <P>/ratio, the <V>/reason_model reading.
• Profile keywords (factors, precedent, plaintiff, defendant, courts, court, decided, outcome, ratio, overruled) are contextual (like map): not reserved outside the profile.

Neighbours and the selection rule

• Holdings and distinguishing are expressible via sources, interpretations, rules and priorities — the profile is not a monopoly but a named superstructure. Selection rule: only the conclusion needed — base forms; binding statuses, constraint, caseBase, agreesWithRules needed — the profile.
• sources vs profile: norm address vs case similarity. An @source anchor on factors/precedent is allowed by the grammar (source_anchor_item in both declarations) and traces the line to the act — as in the teaching package, where every declaration carries @source to decision fragments.
• interpretation vs ratio: the reason model lives in the <V>/reason_model reading with contested status — the result-vs-reason-model choice is made by the reading choice, not by a precedent flag.
• priority vs prefer over: the latter is the author’s declaration of which general rule beats the holding; holding-vs-holding priority of the same vocabulary is not written by the author — the compiler generates it from rank and dates.
• judgment vs distinguishing: a judgment factor without a court answer gives JUDGMENT_PENDING with a working holding, not DISTINGUISHED.
• overruled in a precedent vs edition lifting: the same lifecycle ownership — repeal declared at the repealed, window [decided, T).