Shay c9dfad3017 feat: convergent graph diffusion with exponential-map versor unitization ... Replace the divergent rotation-based diffusion operator with a linear blend + exponential-map re-unitization approach that converges in ~28 steps while maintaining vc < 1e-6. Key changes: - GraphDiffusionOperator now averages neighbors in multivector space and re-projects via per-plane exponentials (cos/sin for rotations, cosh/sinh for boosts in Cl(4,1)) - run_pulse V3: per-token graph topology with input-driven output node, recall via VocabManifold.nearest(), --no-glove flag for compiled pack - Tests updated for V3 API Different inputs now produce different recall rankings from the compiled en_core_cognition_v1 vocabulary, completing Threshold 1 (Semantic Encoding). Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>		2026-05-15 17:02:47 -07:00
..
__init__.py	init: ingest, field, vocab, vault, persona, generate, session layers	2026-05-12 19:14:22 -07:00
operators.py	feat: convergent graph diffusion with exponential-map versor unitization	2026-05-15 17:02:47 -07:00
propagate.py	Tighten hot-path backend consistency	2026-05-15 08:14:38 -07:00
state.py	feat: manifold field topology, graph diffusion operator, vertical pulse	2026-05-15 16:02:48 -07:00