Skip to content

Use Cases

Sanjeev Azad edited this page Jul 4, 2026 · 1 revision

Use Cases

Kuber Map curates FIBO into learner-first, agent-grounding use cases. Each is a hand-picked "core" concept set drawn from across FIBO's domains, with worked examples on every card, a set of validated cross-domain bridges, and an exported context pack an AI agent can ground on.

The five use cases

Use case Core Spans Bridges Pack
Loan origination & underwriting 71 LOAN, FBC, FND, BE 4 export/loan-origination/
KYC & beneficial ownership 58 CMNS, BE, FND, FBC 4 export/kyc/
Securities instruments & issuance 59 SEC, FBC 3 export/securities/
Regulatory reporting & compliance 52 CMNS, FBC, FND, BE, BP, LOAN, CAE, SEC (8 clusters) 4 export/regulatory-reporting/
Derivatives contracts & trading 60 DER, FBC, FND 4 export/derivatives/

284 core concepts, 19 cross-domain bridges in total. On the map these are the default view; the use-case lens (top bar) focuses one, and each detail card shows which use case(s) a concept belongs to.

How a use case is built (spec-driven, grounded)

A use case is a spec under curation/usecases/ — no code change to add one:

  1. Facet spec (<uc>.json) — a list of [local-id, cluster] entries grouped into facets. etl/nominate_core.py resolves every id against the FIBO extract and rejects any that isn't a real class (the anti-hallucination gate).
  2. Bridges (<uc>-bridges.json) — cross-domain relations FIBO doesn't draw natively. etl/bridges.py verifies both endpoints exist and that FIBO doesn't already assert the edge, and requires a rationale + citation for each.
  3. Overlays (<uc>-examples.json, -definitions.json) — worked examples and gap-filling definitions, applied only where FIBO supplies none, marked provenance: curated.
  4. Packetl/export_pack.py emits the grounding closure as pack.json + context.md + an OKF slice.

Regenerate everything with make curate && make pack (see Getting Started).

Why use cases, not just the whole ontology

Dumping all 3,104 FIBO concepts at a learner or an agent is noise. A use case is the minimal coherent slice for a real task (originate a loan, resolve a beneficial owner, file a regulatory report), with the cross-domain links that make it usable — and it's exactly what the value-proof eval measures.

Clone this wiki locally