Version 0.5 (23 May 2026) is an author hand-pass over the v0.4.1 publication-prep draft. The argument, the user-stories analysis, and the references list are unchanged. v0.5 tightens the introduction, hook, and conclusion in the author’s voice; smooths the §6 / §7 prose flow; and keeps the central thesis — audit the nouns before the APIs — as the load-bearing claim. Posted as a publication-prep preprint. The deposit remains implementation-agnostic; EATF appears as a worked example because its source artifacts are publicly inspectable. Version 0.4.1 (18 May 2026) is a minor revision of v0.4-publication (17 May 2026, archived under the same concept DOI). v0.4.1 preserves the abstract, the argument structure, the topology diagram, the temptation scorecard, the distributed-trust design choices, the surveillance-prior-art examples, and the references list of v0.4. It applies the following cleanup so that the public preprint no longer carries internal editorial scaffolding: Removes §9 (Submission strategy), an authoring-stage venue table aimed at the author rather than at the reader of the essay. The remaining sections renumber as §1–§8 followed by Acknowledgements and References, matching the structure promised in §0 of the working draft. No changes to the substantive argument, the four design choices of §5, the open questions in §6, or the general lesson in §7. This revision affects positioning only; the substantive contribution of v0.4 stands. The essay remains a preprint, not peer-reviewed. Publication draft v0.4 (17 May 2026) of On the Crossroads: Marketplace vs. Distributed Trust in Agent Attestation Frameworks. This essay analyzes an architectural choice in agent-attestation frameworks: whether to introduce a central registry for agents, keys, compliance badges, search, reputation, or verification-as-a-service. Using the Agent Evidence Package (AEP) profile and EATF reference verifiers as a worked case, it argues that registry vocabulary can quietly convert an offline-verifiable trust framework into a platform. The recommended alternative is a distributed-trust pattern: public key-history mirrors, operator-managed trust anchors, conformance vectors rather than certification, and no framework-level discovery layer. Version v0.4 adds a related-work paragraph connecting protocol engineering, platform governance, and AI-governance evidence requirements; expands the surveillance-surface argument with OpenPGP keyserver, OCSP, package-registry, Certificate Transparency, and Rekor examples; and adds a submission strategy distinguishing Zenodo-first release from later arXiv or venue submission. Tyche Institute is a research entity, not a trust service provider or qualified trust service provider. EATF is referenced as an open specification and reference implementation, not as an eIDAS trust service or compliance certification product.
Anton Sokolov (Sat,) studied this question.