First archived release (Zenodo DOI target). Cognitive self-state channels (memory, affect, time, ethics, identity, continuity) injected per layer into a frozen Gemma 4 E2B adapter at smoke scale. Headline — Pillar A (conflict / override): under direct conflict, the channel overrides contradicting text 264/265 (98–100%) across affect/time/ethics — emergent (never a training objective) and pre-registered (docs/english/15). Also: 6/6 causal + integrated as one organ (no interference); channel-vs-text P1 (channel > faithful text on 4/6, zero-shot); B-1 token-cost (0 vs ~166 tok/turn); B-2 persistence null-at-smoke (reported honestly). Honest scope: this is an adapter on a frozen base at smoke scale — NOT a from-scratch native architecture. The native experiment (channels from layer 1) is specified (docs/english/16) and is the decisive next step. License: GPL-3.0 (code) + CC BY-SA 4.0 (docs).