nx_verdict_phrase.nx

nx_verdict_phrase.nx -- human-legible phrasing for grader verdicts. CAPABILITY_COMPLETENESS: FULL (for shipped tier-axis verdicts; new graders register their phrasing via nx_verdict_phrase_set) User direction 2026-05-16: "also make sure the math is human legible like the skin is too plastic or something even more explanatory or she is 50 percent asian in this pic etc" The substrate's tier-axis verdicts are sealed integers + Q10 scores -- great for the closed-loop trainer + ledger, useless for a human reading the report. This primitive bridges the machine-readable verdict world with the human-readable explanation world. Two phrasing modes: MODE_TERSE -- one short clause, suitable for batch dashboards: "skin too plastic" "background fragmented" "vein signal absent" MODE_VERBOSE -- one full sentence with the score reading: "Skin texture verdict PLASTIC_SHEEN at energy 0.83 (expected 0.4-0.7 for natural skin); over-smoothed render." Phrases are emitted as NUL-terminated UTF-8 byte sequences. Caller passes a buffer of NX_VP_MAX_PHRASE_LEN bytes; primitive writes the phrase and returns the byte count written. CONTINUOUS-CLASS PHRASING (e.g., "30% east-asian / 50% european / 20% african" -- the user's "50 percent asian" example): the phrasing is done via NxVerdictBlend struct (sealed-enum class list + per-class Q10 weight) -> phrase rendering with thresholds (drop classes below 5% Q10, sort descending, format "X% kind / Y% kind / Z% kind"). genealogy_id: substrate_legibility_layer_2026_05_16 lineage_id: nx_verdict_phrase_v1