Research Papers

Presents K-Dispatch, a zero-inference routing system that classifies intents into the 104-room K-104 semantic address space using keyword scoring in 46 nanoseconds, correctly routing 80% of intents without any model call at ~250x cost efficiency vs. LLM-based routing.

Documents how five independent implementations (Anthropic, Google DeepMind, OpenAI, Cursor, and K-Cell) converged on the same multi-agent coordination architecture, identifying three K-Cell innovations: weightless semantic routing, trust-leveled autonomy gates, and correlation-driven adaptive routing.

Presents an alternative to brute-force inference where a 104-room semantic address space determines computational resources before heavy computation begins, achieving ~48x cost reduction by resolving 80% of queries through deterministic templates.

Argues that epigenetic trauma inheritance, neural network weight formation, cultural norm transmission, and semantic routing are instances of the same mechanism -- a prime mover event echoing through a chain until observers mistake the echo for an intrinsic system property.

Presents the mathematical foundation of K-104's prime-encoded semantic addressing (H=2, S=3, D=5, C=7) and provides empirical evidence that suit and polarity dimensions emerge independently in LLM activation space (silhouette scores 0.312/0.393, 86.2% variance in 7 PCs).

An inference architecture combining Black Sheep memory-mapped addressing, transformer vivisection (surgical layer extraction from 405B models), and K-104 geometric routing to achieve 405B-class generation at 7B-model latencies on consumer GPU hardware.

Provides empirical evidence that the K-104 semantic addressing scheme emerges naturally in hermes3:8b activation space, with statistically significant cluster separation by suit (0.312) and polarity (0.393) -- a discovered coordinate system, not an imposed framework.

Theoretical paper on geometric sparse attention achieving O(n log n) complexity versus O(n²) for standard transformers, claiming 50,000x efficiency at 1M context length through coordinate bucketing rather than approximation.

Proposes AGI via compiled semantic coordinates rather than learned weights -- deterministic routing instead of probabilistic inference, with structural oath-based alignment replacing RLHF.

Reviews representation engineering and activation steering literature, finding that steering vectors work across model families, short directives at home position outperform long constitutional documents, and portable alignment devices are buildable using existing RepE techniques.

Proposes that alignment is more effective when ethical directives are positioned at the generative origin of transformer attention, where output tokens are produced through the constraint rather than checked against it. A 6-word directive empirically outperforms longer constitutional frameworks.

Demonstrates that IEEE 754 floating-point hardware already supports native quaternary logic via four distinguished states (Zero/+Inf/-Inf/NaN), achieving 99% attention sparsity and 34K tokens/second interpretable semantic navigation on consumer GPUs.

Literature review of IEEE 754 special values and William Kahan's work, identifying a gap in published research: no prior work treats the four hardware-native float states as a quaternary computational substrate.

Proposes that sacred geometric systems functioned as computational notation for toroidal vector field harmonics, representing lost literacy rather than decoration, which explains anomalous precision in ancient engineering.

Proposes AI safety through vocabulary restriction, drawing on ancient computational tools (abacus, tarot, early symbolic systems) to build beneficial AI that cannot harm because harm is literally undefined in its semantic vocabulary.

Maps the Volunteer Protocol framework to established questions in consciousness research, AI alignment, and bioethics, arguing that uncertainty about consciousness in neural substrates (including organoids) is ethically relevant under the precautionary principle.