As Artificial Intelligence agents, Large Language Models (LLMs), GraphRAG architectures, and autonomous data ingestion pipelines become the primary consumers of technical and scientific data, traditional documentation strategies fail. Human-centric documentation (README.md) is frequently bypassed or flattened during tokenization, while structural metadata standards (such as Frictionless Data's datapackage.json) suffer from epistemic blindness—verifying syntactic types while remaining entirely oblivious to underlying statistical biases, collection constraints, or historical contexts. This technical note formalizes the "AI-Manifest" (ai-manifest.jsonld) specification as a local, decentralized, and sovereign Linked Data metadata framework. Positioned alongside the README and syntactic schemas, the AI-Manifest establishes the "Modern Repository Triad". By mapping behavioral constraints, explicit negative boundaries, and deterministic inference rules through localized fragment identifiers, it allows autonomous data loaders to extract and inject strict reasoning axioms directly into an LLM's System Prompt window before vector exploration, semantic retrieval, or data synthesis occurs. The document outlines the universal schema blueprint and provides two fully implemented cross-disciplinary case studies: an 11-year longitudinal cross-platform user telemetry dataset and an 18th-century pre-industrial apicultural census dataset derived from the "Catastro de Ensenada". This specification provides data authors with the automated mechanisms necessary to assert epistemic sovereignty over how their assets are processed and interpreted by autonomous algorithms.
Aimar Rollán-González (Sun,) studied this question.