Not slides — working software. Requirements expressed as proofs instead of documents, licences verified against live public registries, documents classified by AI at upload, screening that only flags at very high confidence, and progressive caps so customers move money while the paperwork catches up. Every surface below reads and writes one canonical customer spine.
The full KYB flow on the canonical proof model: per-country document registries, live CNPJ and FINTRAC checks, Claude classification of every upload (PDF-aware), UBO passport verification, attestation, and an integrator's JSON view.
Step 2 of the journey: $10 payout tests on Fedwire, SWIFT, SEPA, FPS, PIX, SPEI and HKD FPS — Codex-funded — plus onramp settlement instructions with reference-based reconciliation.
What the analyst sees while the customer types: gates satisfying in real time, evidence per proof, and RFI drafts generated from whatever is still open.
RFIs generated from open gates — never written by hand. Magic-link delivery, reminder cadences, customer uploads run through the same classifier, items auto-close as proofs satisfy.
Proof-of-payment in, stablecoins out. Claude parses MT103s and wire confirmations, duplicates are hard-blocked, facilities meter utilization, and repayment rails carry per-chain addresses plus USD wire where the memo is the reconciliation.
The same machinery as an API, written in the Codex FX idiom: create customers with inline registry enrichment, synchronous document classification, submit for progressive decisions. Custom docs with guides, a glossary and a live sandbox.
Approve a customer in Codex Zip, through the hosted form, or over the API — the record lands in the same canonical registry. The compliance console sees it, the RFI manager references its open proofs, and the prefunding desk sizes its facility. That's the argument the suite makes: onboarding isn't a form, it's a shared state machine.