nx_write_exceed_gate.nx

nx_write_exceed_gate.nx -- WRITING arc EXCEED-BENCH (W-EXC-1): the MEASURED S-class-exceed verdict for nishi-write, in the codebase's competitive-census style (exceed is MEASURED, never asserted -- STARTUP_BRIEFING law 5). A 10-capability matrix. Each NISHI cell is PROVEN by actually RUNNING the real organ in this gate (no cell is declared true without a passing run). The incumbent columns (Hemingway / Grammarly / ProWritingAid / Scrivener) are STATED PUBLIC capability -- a named reference yardstick, NOT a fabricated score (the same discipline as the WMS-R11 "git as yardstick" exceed-bench). caps 0-4 (SHARED): readability-grade, adverbs, passive, weak/qualifiers, hard-sentence -- incumbents have these; Nishi PROVES parity. caps 5-9 (NISHI-UNIQUE): scene-extraction, character-presence, descriptor/ image-gen profile, register-routing+governance, integrated sovereign stack -- the incumbents lack these. EXCEED = nishi_cov (proven by running) STRICTLY GREATER THAN the best incumbent, with the 5 shared cells at proven parity and the 5 unique cells the incumbents do not cover. NO-FABRICATION control: an honestly-absent capability (grammar/spell-check, not yet built) stays 0 -- the census reflects reality. Evidence -> stdout + knowledge/status/write_exceed_gate.log. Exit 0/1. Sovereign x86_64. license_tier: ORIGINAL