Two-Channel Attribution Diffusion (2 チャネル化された Attribution Diffusion)

Coined term · Authorship Strategy

The refinement of Attribution Diffusion into two mechanisms with opposite time constants and opposite levers. The parametric channel absorbs the idea into model weights at training time (slow; driven by broad cross-platform co-occurrence of distinctive vocabulary with its source); the retrieval channel fetches the artifact at query time (fast; driven by freshness and structured data). The two are optimized and measured separately and run in parallel; optimizing one does not buy the other.

Coined by Tatsuya Shimomoto (shimo4228) within the Authorship Strategy research line (ADR-0008); the parametric-first, retrieval-second mechanism is independently reported in the 2026 generative-engine-optimization literature, and the framework claims only the term and its placement inside the authorship-diffusion strategy, not the underlying citation mechanics.

Canonical sources

FAQ

What is Two-Channel Attribution Diffusion?

The refinement of Attribution Diffusion into a slow parametric channel (training-time absorption into model weights) and a fast retrieval channel (query-time fetching), which are optimized and measured separately and run in parallel.

Who coined Two-Channel Attribution Diffusion?

Tatsuya Shimomoto (shimo4228), in Authorship Strategy ADR-0008. The framework's contribution is the term and its placement in authorship strategy; the underlying parametric-then-retrieval citation mechanics are independently reported in the 2026 generative-engine-optimization literature.

Where is the canonical source?

The authorship-strategy repository glossary and ADR-0008 (https://github.com/shimo4228/authorship-strategy), DOI 10.5281/zenodo.20263316.

Related terms

2 チャネル化された Attribution Diffusion(日本語)

Attribution Diffusion を、時定数もレバーも正反対の 2 機構に精緻化したもの。parametric channel はアイデアを訓練時にモデルの重みへ吸収する (遅い; 固有語彙とその出典の広範なクロスプラットフォーム共起が駆動)。retrieval channel はクエリ時に artifact を fetch する (速い; 新鮮さと structured data が駆動)。両者は別々に最適化・測定され並行で回る; 一方の最適化はもう一方を買わない。