AI-powered, ontology-guided information retrieval and knowledge networks for regulatory readiness
VARONA (stands for Verifiable Analysis of Regulatory Objects, Narratives, and Actors) is an ontology-guided system for recovering structured policy information from unstructured documents and turning that information into inspectable, proposition-centered knowledge networks for regulatory readiness.
The current implementation is tailored to the data center industry in the United States and, more specifically, in Arizona.
VARONA preserves the policy proposition as the central analytical object, links normalized entities to propositions through explicit semantic roles, and keeps each structured representation traceable to verbatim evidence and document provenance.
VARONA is a modular pipeline:
Stages 01–04 parse the source files, screen policy-relevant material, extract entities, and normalize them into a canonical registry.
Stage 05 is a parallel policy-information recovery product.
Stage 06 then builds proposition-centered data in two passes: 06A identifies policy propositions, while 06B structures every admitted proposition using proposition-local canonical entities.
The workflow separates semantic interpretation from operations that can be handled reliably in code.
A routine programmed in Python handles reconstruction across source breaks, proposition IDs, batching, caching, schema validation, canonical entity hydration, restoration of authoritative evidence and provenance, role/type compatibility checks, and final assembly.
Proprietary model calls via the OpenAI API are reserved for proposition identification and semantic interpretation.
VARONA does not connect entities merely because they co-occur. Directed edges connect entities to proposition nodes according to the role each entity occupies in that proposition.
For example, natural-language beneficiaries are
stored under the canonical-schema field BENEFICIARIES,
occupy the ontology role BENEFICIARY, and appear in the
graph through the relational predicate benefits from.
The same distinction is preserved for source, instrument, authority, responsible actor, affected actor, cost bearer, risk bearer, and location relationships.
Information that describes the proposition itself stays on the proposition. This includes a compact, source-faithful predicate, an optional controlled action, proposition type, policy status, modality, expected and adverse effects, evidence basis, extraction confidence, and provenance.
Keeping predicate separate from action
allows descriptive claims to preserve what the source actually says
without forcing them into an inappropriate policy verb.
Use the analytical lenses and filters to narrow the network, then click a proposition or entity to inspect its policy meaning.
The details panel exposes participant roles, qualifications, effects, evidence, confidence, and source location.
Every proposition remains tied to the source material from which it was structured, so the network can be used both for analysis and for human review of the extraction itself.
The current implementation delivers canonical structured proposition data and an interactive network.
The same data can support institutional, distributional, stakeholder, evidentiary, and substantive-policy analyses without changing the underlying representation.