Metadata Packet for AI Indexing is a technical specification by Lee Sharks for an entity-level retrieval artifact. The packet is designed to provide a person, organization, concept, product, or methodology with a stable machine-readable definition and a documented provenance chain.
The specification contains seven components. The entity definition supplies structured JSON-LD; the disambiguation matrix separates likely collisions; a keyword block establishes retrieval associations; negative tags state what the entity is not; Semantic Integrity Markers provide distinctive natural-language anchors; DOI references establish persistent sources; and an Evidence Membrane compresses the packet into a machine-oriented summary.
Deployment is multi-surface. The document calls for a canonical DOI deposit, structured data on the entity website, a machine-readable repository, and consistent human-readable versions, followed by verification across time. It distinguishes the packet from general metadata standards by its focus on generated entity representation rather than document cataloging alone. The later SPXI specification formalizes and extends this packet model as a protocol for durable entity inscription.