CITATION.cff

cff-version: 1.2.0
message: "If you use Shannon-Prime, please cite it."
title: "Shannon-Prime: a two-ring memory architecture for long-context transformers"
authors:
  - family-names: Daniels
    given-names: Ray
    alias: KnackAU
abstract: >
  A memory architecture that attaches to a frozen pretrained transformer:
  query-directed sparse recall (a ±1 Rademacher projection router), attention-sink
  pinning to preserve quality under sparsification, a two-ring design offloading the
  cold KV cache to byte-addressable storage, and O(N) selection. Proof-of-mechanism
  on Qwen3-0.6B; retrieves at 32k context with the resident KV cache held to ~8 MB.
keywords:
  - long-context inference
  - kv cache
  - sparse attention
  - retrieval
  - transformer
version: 0.1.0
date-released: 2026-06-03
license: MIT
repository-code: "https://github.com/nihilistau/Position_Is_Arithmetic"