nx_term_registry.nx

nx_term_registry.nx -- open string-id pool with sightings + stage. Foundation of the LEARNING ingestion layer. Replaces hand-written sealed-enum dictionaries (nx_outfit_vocabulary, nx_location_dict, etc.) with a self-grown registry that LEARNS its vocabulary from the ingested corpus. Every noun phrase the discovery layer finds gets registered here with provenance: where first seen, how many times sighted across the corpus, what semantic role (outfit / location / character / action / object), and which self-surfacing-intelligence stage it has reached. Per cardinals: feedback-self-surfacing-intelligence-staged-autonomy: STAGE1 SUPERVISED (default for fresh sightings), STAGE2 PROMPTED (>= N sightings + 0 contradictions), STAGE3 UNSUPERVISED (user-pinned or arc-anchor confirmed). feedback-user-owns-every-bit: this primitive holds the LIVE counts in-memory; durable snapshots ship to nishi-library/seeds/*.toml on demand, never via background daemon. feedback-loras-and-negatives-are-patches: the registry is a substrate-native CLOSED-LOOP measurement surface -- not a black-box embedding -- so confidence is auditable per-term. nx_safety_envelope: intended_use: "Open string-id pool for discovered noun phrases, with sighting counts, semantic role tags, and self-surfacing stage." sil_target: SIL2 asil_target: QM dal_target: DAL C iec_62304_class: NONE evidence: [no_floating_point, fixed_capacity_hash_table, bounded_probe_length_per_jpl_rule_2, sealed_role_enum, sealed_stage_enum,