TBC Researchhttps://tbcresearch.org/Open research and results from TBC Research.Design note: the shadow of a higher-dimensional gridhttps://tbcresearch.org/research/unit-distance/https://tbcresearch.org/research/unit-distance/Why this site is drawn from the 2026 disproof of Erdős's unit distance conjecture, and what the moving lattice on the home page actually computes.Thu, 08 Oct 2026 12:00:00 GMTAn autonomous overnight loop for Claude Codehttps://tbcresearch.org/research/nightcrawler/https://tbcresearch.org/research/nightcrawler/Splits a mission into fresh-context Claude Code episodes with structured handoffs, per-episode budget caps, and git-diff checks of claimed progress.Hebbian procedural memory for coding agentshttps://tbcresearch.org/research/brainbox/https://tbcresearch.org/research/brainbox/Learns which files are accessed together, which errors lead to which fixes, and which tool chains recur, then recalls them before the agent searches.Hybrid Neural Engine, GPU, and CPU LLM inference on Apple Siliconhttps://tbcresearch.org/research/ane-infer/https://tbcresearch.org/research/ane-infer/A from-scratch engine running Qwen3.5-2B across the Apple Neural Engine, Metal GPU, and CPU; Metal decode matches llama.cpp at about 32 tok/s.A batch-1 INT4 decode megakernel on a 2018 Turing GPUhttps://tbcresearch.org/research/latticemk/https://tbcresearch.org/research/latticemk/A GPTQ INT4 decode megakernel on a Quadro RTX 4000 running 1.09–1.64× faster than llama.cpp Q4_0 on Qwen3-0.6B and 1.7B.Speculative speculative decoding on an M4: a documented ceilinghttps://tbcresearch.org/research/ssd-apple-silicon/https://tbcresearch.org/research/ssd-apple-silicon/Can SSD-style parallel speculation beat plain MLX speculative decoding on Apple Silicon? No: 66.9 tok/s is the practical ceiling, set by memory bandwidth.A real-time neural synthesizer on the Apple Neural Enginehttps://tbcresearch.org/research/ane-synth/https://tbcresearch.org/research/ane-synth/Neural audio inference directly on the Neural Engine, bypassing CoreML: about 157 µs per 8-voice buffer, roughly 79× real-time headroom.Measuring and repairing a checkpoint's surgery reservehttps://tbcresearch.org/research/theseus/https://tbcresearch.org/research/theseus/Two transformers with identical outputs can respond very differently to quantization, LoRA, or merging. Theseus detects and repairs the difference.Self-stabilising, auditable brain-computer-interface decodershttps://tbcresearch.org/research/neural-decompile/https://tbcresearch.org/research/neural-decompile/Treats decoders as programs: diagnoses drift, repairs from about 30 s of data with label-free correction, and compiles to dependency-free streaming Rust.A compositional morpheme tokenizer built on holographic reduced representationshttps://tbcresearch.org/research/morph-hrr/https://tbcresearch.org/research/morph-hrr/Each word becomes one fixed-width vector binding prefix, root, and suffix with circular convolution, so morphemes can be algebraically recovered.