nx_ingest_scene.nx

nx_ingest.nx -- WRITING arc, rung W-ING-1 (X-WRITE-ING-001): the content-agnostic SCENE / ENTITY EXTRACTOR -- the ingestion point that turns generated prose into reusable structured DATA for the companion (elder-ai) + image-gen. Per the 2026-06-12 fiction-ingestion seam, this is engine PLUMBING: register-blind, so the SFW lane and the operator's local-LLM explicit lane run the SAME organ. ROSTER-BASED + DETERMINISTIC (no LLM in the loop): the caller supplies the character bible (a packed list of names); the extractor finds, per scene, who is present and how strongly. That is exactly what image-gen needs (render the characters present; the focus character is the subject) and what the companion needs (who is in which scene, dialogue density). REAL-TIME == PRIOR: the extractor is a pure function of (text, roster). Running it on each scene as the LLM finishes it (streaming) yields byte-identical scene records to running it once on the whole work (batch). The gate proves this. Scene break = a blank line (>=2 consecutive '\n'). Character match = WHOLE-WORD, case-insensitive ("Yuki" does NOT match inside "Yukiko"); a trailing apostrophe is a boundary so possessives ("Kenji's") count as a mention. Marker breaks (***, # heading) and a descriptor/outfit/setting lexicon are the next layers (DATA-driven, rule 25) and ride on this same per-scene structure. out[] layout, per scene s (stride = 5 + nnames; caller allocates >= maxsc*(5+nnames) i64): out[base+0] = char start (absolute in the buffer passed) out[base+1] = char end (exclusive) out[base+2] = dialogue segments (count of double-quote pairs) out[base+3] = focus character index (most mentions; -1 if none present) out[base+4] = total mentions across all roster names out[base+5+k] = mentions of roster name k ig_extract returns the scene count (capped at maxsc). Pure integer, NO syscalls. Reuses nx_writecraft for char classification (DRY). license_tier: ORIGINAL module: nishi-core.write.ingest depends: nishi-core.write.craft capability: WRITE_SCENE_EXTRACT