nx_noun_phrase_scan.nx

nx_noun_phrase_scan.nx -- bounded heuristic noun-phrase discovery. Walks a source byte region looking for STRUCTURAL English cues (determiners + key verbs + prepositions) that precede a noun phrase, then captures the next 1-4 token span as a candidate. Cues are GRAMMATICAL not TOPICAL: "wearing", "in the", "at the", "her", "his" -- not "apron" / "kitchen" / "bedroom". The actual vocabulary is discovered from what follows the cues. Output: caller-allocated array of PhraseCandidate records. The term registry observes each candidate and grows the open vocabulary. No closed-set dictionary anywhere in the pipeline. Per cardinals: feedback-loras-and-negatives-are-patches: structural cues, not topical lexicons -- the substrate measures statistical co-occurrence rather than memorizing seed lists. feedback-self-surfacing-intelligence: scanner emits CANDIDATES with confidence; registry grades them through STAGE1/2/3 over time. feedback-bounded-loop-discipline-jpl-rule-2: every loop carries explicit BUDGET ceiling. nx_safety_envelope: intended_use: "Surface noun-phrase candidates from a beat's source byte range using grammatical-cue heuristics, with role-hint + confidence." sil_target: SIL2 asil_target: QM dal_target: DAL C iec_62304_class: NONE evidence: [no_floating_point, no_closed_topical_dictionary, bounded_outer_and_inner_loops, out_array_capacity_respected] hazard_register: [bug-tape-cue-pattern-false-positive, bug-tape-phrase-span-overruns-sentence] residual_risk: "Structural-only English cues today; non- English text gets fewer candidates emitted