COHERENCE HEAD

Keeps the output language consistent on multilingual input — stops the model drifting into the wrong language mid-answer. It weights attention toward tokens in the target language. Conceptual demo on tagged tokens.
AMBER
2D · Quantitative Channel
CANVAS 2D
demo: sine — replace in draw2D()
CANVAS 2D UNAVAILABLE
3D · Spatial Channel
SOFT-GL · NO GPU · NO LIB
DRAG ROTATE · WHEEL ZOOM
demo: lattice + orbit — replace in build3D()/draw3D()
RENDER FAILED —
Notes

Coherence head. On multilingual input, this head keeps the generated output in the target language, preventing the model from slipping into the wrong language partway through a response. It does so by preferentially weighting attention toward same-language tokens. Amber tier: shown conceptually on language-tagged tokens; the within-language weighting is the documented behavior, the specific weights are illustrative.