"SFH: A Minimal Semantic Layer for the Web (One File, Zero Dependencies)" (github.com)

A public draft proposes the Semantic Stack: an external semantic layer and a tiny JSON‑LD “Deterministic First-Hop” (DFH) descriptor served at /.well-known/stack that gives AI systems a stable, public “start here” for any topic. The idea addresses a core cause of LLM hallucination — the web lacks universal, public-facing topic anchors — by prescribing one canonical root per concept (represented by a predictable five‑domain anchor set), plus mirrors for plural, category and contextual expansions. The proposal is explicitly complementary to RDF/OWL and existing LOD resources, not a replacement; it functions as an external entrance point before internal graphs and reasoning engines are used. Technically, each DFH document names a rootTopic, five anchor URIs (type/entity/url/sitemap/canonical), mirrors, owl:sameAs links, and optional SPARQL endpoints or DCAT/VoID metadata. Discovery workflow: clients fetch /.well-known/stack to deterministically disambiguate homonyms (e.g., jaguar-as-animal vs jaguar-as-car), then federate into internal graphs with tools like Comunica or Jena. The spec includes freshness signals (ETag/feeds), decentralized governance similar to DNS (competing roots allowed), and a roadmap: v1 anchors+DFH, later disambiguation/change feeds, then a separate Verification Stack to attach provenance and veracity — together forming a Universal Knowledge Stack to reduce hallucination, drift, circular citation, and improve grounding for AI systems.