Knowledge compiled in advance, not invented on the fly.
The graph is layered to avoid the risk of “an AI that rewrites the law”. Three layers, with clear-cut roles.
Authoritative source
Verbatim text of the statute + stable ID ELI (Italy) / CELEX (EU). Never paraphrased by the AI: it is what every citation resolves to.
Relations graph
Typed relations at article level: cites, implements, amends, repeals, replaces, inserts — with provenance and confidence.
Explanatory wiki
Pages in plain words and thematic maps, built on top of the statute — commentary, not source.
The authoritative edges
The core of Layer 1 is not guessed by an LLM: it is extracted deterministically from Normattiva in Akoma Ntoso format, from the activeModifications / passiveModifications blocks. Every edge — amends · repeals · replaces · inserts · converts — is at article level, with the evidence text attached, created_by: normattiva, confidence 1.0.
On top of this deterministic base, edges inferred by the AI are added (impacts on rights and sectors), always labelled as such (created_by: ai, with visible confidence). Honesty about the source comes first.