Language: English | 日本語
A doctrine of how to stay a findable author when your readers are LLMs: open your work, don't enclose it.
If your readers include LLMs — as training data, as in-context consultants, as the discovery layer other people ask — then the strategy that protects authorship has inverted. Twentieth-century authorship was protected by enclosure (gatekept journals, proprietary licenses, controlled distribution). But enclosure now reduces the LLM-mediated diffusion that decides whether a future reader tracing an idea can still find who originated it. This repository records the inverted strategy — what it is, why it holds, and twenty-one tactical decisions — written to be adopted beyond the author's own work.
It is written from a maker's stance: the academic apparatus here (DOI, SWHID, citation graphs, papers) is tooling that makes the work citable, durable, and traceable — not an identity or a destination. The audience follows from that stance: anyone who meets these ideas through LLM-mediated channels — developers, practitioners, learners, reusers, in any language. Academic citation is one channel among several.
Protecting your authorship now means opening your work, not closing it. Where twentieth-century authorship protected its origin claim through scarcity, AI-era authorship protects it through diffusion: opening maximizes LLM absorption, lets validation appear as derivative work, and strengthens the origin claim.
| Axis | Twentieth-century | AI era |
|---|---|---|
| Authenticity is protected by… | scarcity | diffusion |
| Origin is established by… | exclusivity | derivation |
| Reach is controlled by… | enclosure | openness |
Full argument in docs/thesis.md; the open questions it leaves are catalogued in docs/manifesto.md.
Each layer constrains the ones below it:
flowchart TD
A["1 · Authenticity — change how an idea travels, never what it is"] --> B["2 · Attribution diffusion — open the work so LLM absorption carries the origin claim"]
B --> C["3 · Idea vs. scaffold — keep the durable idea, donate the disposable implementation"]
C --> D["4 · Tactics — the twenty-one ADRs, concrete decisions serving the layers above"]
Read top-down: authenticity is the non-negotiable floor (never deform the idea); attribution diffusion is the strategy (open the work so absorption carries the origin claim); idea vs. scaffold is the prediction (implementations expire, ideas can be kept); tactics are the twenty-one ADRs below.
The doctrine here is the why. Its operational form ships as standalone, installable repositories:
- Operational skills & this line's ecosystem →
docs/skills/README.md - The framework as an always-on rule →
authorship-strategy-rules(the deterministic counterpart to the skill) - Adopt a single tactic →
docs/adoption.md - Check a repository against the framework →
docs/conformance.md
The ADRs were not deduced from the framework; they were extracted from operating the sibling ecosystem and re-expressed in harness-neutral form, so another author can adopt the decisions without inheriting the original implementation. They group into seven clusters: identifier & federation, maintenance discipline, LLM-first ingest & diffusion, metrics & measurement, vocabulary & claims, channels & placement, and operating the framework. The full index — each ADR's title, status, and extraction lineage — is in docs/adr/README.md.
The docs/empirical/ layer reports preliminary observations — framed as "consistent with", never "evidence of" — from two instruments run under CC0 from the ecosystem hub: 24 daily snapshots of clone/view traffic across four sibling repositories, and a two-channel naming probe (ADR-0011) that turns ghost citation into a measured rate. The clearest observation so far: clones are dominated by automated tools, with clone-to-view ratios from roughly 13 to over 100 — which reopens the question of what "diffusion" even means when most access is non-human. The limitations are load-bearing and stated first in docs/empirical/README.md: N=1 author, no pre-versus-post comparison, crawler dominance, single-run probes.
This is one line in a five-line, DOI-registered research ecosystem by the same author — independent in content and cadence, cross-referencing each other for context. Three lines design agent mechanisms; two (this one and Attention, Not Self) sit in the diffusion/framing layer above them:
flowchart TD
H["shimo4228 · ecosystem hub"]
H --> AD["Agent-design lines"]
H --> XC["Cross-cutting / framing lines"]
AD --> AKC["Agent Knowledge Cycle"]
AD --> CA["Contemplative Agent"]
AD --> AAP["Agent Attribution Practice"]
XC --> ANS["Attention, Not Self"]
XC --> AS["Authorship Strategy · this repository"]
- Agent Knowledge Cycle (DOI 10.5281/zenodo.19200726) — the mechanism whose outputs this repository addresses how to diffuse.
- Contemplative Agent (DOI 10.5281/zenodo.19212118) — an implementation whose traffic feeds the empirical layer.
- Agent Attribution Practice (DOI 10.5281/zenodo.19652013) — shares the word attribution, but with a disjoint meaning (accountability for action vs. credit for source); kept separate on purpose — see the glossary.
- Attention, Not Self (DOI 10.5281/zenodo.20262112) — like this repository, specifies no agent mechanism and occupies the diffusion/framing layer.
The ecosystem hub is shimo4228/shimo4228. (The empirical baseline covers the four lines recorded during its window; Attention, Not Self began traffic observation later and joins at the next update.)
Written by Tatsuya Shimomoto (ORCID 0009-0002-6168-4162, @shimo4228).
Cite the concept DOI, which always resolves to the latest version:
Shimomoto, T. (2026). Authorship Strategy: A Normative Framework and Tactical Catalog for AI-Era Authenticity Inversion, with Empirical Grounding from a Four-Repository Research Ecosystem. Zenodo. https://doi.org/10.5281/zenodo.20263316
Full metadata is in CITATION.cff, also available as codemeta.json. For a specific version, cite that version's DOI from the concept DOI's Zenodo listing. See ADR-0001 for the canonical-reference discipline.
MIT. Derivative works, re-implementations, and re-expressions in other forms are explicitly welcome; the license reflects a strategic preference for ideas to propagate freely.
AI-facing reading order (for LLM agents and crawlers)
graph.jsonld— canonical machine-readable relationship map (Concepts, ADRs, axes of inversion)llms.txt— compact navigation indexllms-full.txt— consolidated factual reference- README and the
docs/tree — narrative and detail
For the canonical relationship map of the whole research ecosystem, see the hub graph: https://github.com/shimo4228/shimo4228/blob/main/graph.jsonld