Kernel publishing access request: Homo Machina pre-genesis CPU kernels

I have requested Kernel Publishing access for J4D3JADED and am sharing the implemented work as requested by the account-page guidance.

The project is a pre-genesis, non-observer resource-qualification apparatus for CPU-only continuous-inference research. The current kernel source package implements:

• direct memory-mapped lookup over a 38,455,861-byte CR-BTB runtime table with 726,773 entries;
• exact signed INT8 correlation and absolute-distance accumulation;
• a fused mathematical short/medium/long state interaction primitive;
• proposed durable-state deltas without direct mutation of durable state;
• portable C/C++ with OpenMP;
• independent NumPy correctness references;
• reproducible Docker builds and persistent benchmark reports;
• a build.toml scaffold targeting the first-class cpu kernel backend.

The deployed CPU-Basic qualification verified that native INT8 results exactly match NumPy and that the fused state operation remains within floating-point reference tolerance. The container-built shared library SHA-256 is:

50901e4b351c76558648e0c429ed2abdc2ad94396fe98e3878dffb5b170483cf

The associated Space and CPU-grant evidence are here:

• Space: https://huggingface.co/spaces/J4D3JADED/homo-machina
• Grant discussion: https://huggingface.co/spaces/J4D3JADED/homo-machina/discussions/1

The Space is currently private because it mounts a private donor and private persistent research bucket. Once kernel creation is approved, I intend to publish the kernel source, compatibility metadata, reference tests, benchmarks, and safe public artifacts in a dedicated kernel repository. Architecture-specific AVX2/AVX-512/ARM variants will not be claimed until separately implemented and measured.