Plan, apply and roll back changes to a production ontology, with a blast radius report and a proof an auditor can re-check.
Rust ai-nativedescription-logicsformal-verificationinfrastructure-as-codeisabelle-languageknowledge-graphleanlean4mcpmcp-serverontologyowlowl2reasoningrustshaclsparqltheorem-proving
export interface Document { id: string; title: string; text: string }
export interface Claim { document: string; predicate: string; object: string }
export interface Contradiction {
id: string
subject: string
// 'provenance-split' and 'typing': two documents type the same individual
// incompatibly. 'disjointness': a reasoner-caught axiom violation.
// 'conformance': a claim contradicted by the artifacts published alongside
// it, e.g. a README asserting a standard the corpus does not exhibit,
// established by a validator run rather than the document-to-ontology
// pipeline. Every kind still cites real documents in its claims.
kind: 'provenance-split' | 'disjointness' | 'typing' | 'conformance'
claims: Claim[]
}
export interface GraphView {
classes: { iri: string; label?: string }[]
properties: { iri: string; label?: string }[]
edges: { source: string; target: string; warrant?: Warrant }[]
}
/**
* Where an edge came from, and what is known about it.
*
* `asserted` someone wrote it down. Nothing is claimed beyond that.
* `certified` derived, AND the Lean checker accepted the derivation.
* `rejected` a derivation the checker REFUSED. Exit 1, with the rule named.
*
* There is deliberately no `unproved`. Measured on ies-core, 151 of 151
* materialised inferences carried an accepted certificate, so a category for
* "derived but unvouched" would have been drawn rather than observed.
*
* Absent means `asserted`, so a source that knows nothing about certificates
* renders exactly as it did before.
*/
export type Warrant = 'asserted' | 'certified' | 'rejected'
export type Decision = { kind: 'accept' | 'reject'; note?: string }
export interface Chunk { type: 'text' | 'tool_call' | 'unscripted'; value: string }
export interface DemoSource {
corpus(): Promise<Document[]>
graph(): Promise<GraphView>
findings(): Promise<Contradiction[]>
resolve(id: string, decision: Decision): Promise<void>
ask(question: string): AsyncIterable<Chunk>
}