You have the data, but can you see it?

One relationship, as Round Table stores it

The risk you never saw and the opportunity you never spotted are the same failure.

See it on public dataBring us your question

Instead of an issuer saying "it's backed, trust us", each claim names its evidence: the backing attestation and who signed it, the holder, the recorded history and the current permission.

Live graph, public data, as at 2 October 2026:

Every edge carries its own source, basis, confidence and the date it took effect, which is the difference between a knowledge base and a diagram. Open any of them yourself.

What public sources alone allow

Live, public data only. We pointed Round Table at the AI sector using nothing but public sources.

That is what public sources alone allow. On your own assets, Round Table works from your own records rather than inferring from public ones, so each claim traces to a source you hold.

See it on public data

What is judgment here, and what is sourced

The claims in this graph come from sources you can open. The weighting applied over them is Shayne Heffernan's: which entities sit upstream, which are single points of failure, which findings rank as critical, and how far each claim's evidence is trusted. Those are judgments. They are where most of the analytical value sits, and they are attributed to him rather than presented as facts. Where a reading of the graph is ours rather than a source's it is marked derived and carries no source URL, because inventing one would be the failure this map exists to prevent.

Why each part of the market needs this layer

Everyone in a market is asking a version of the same question. What is this really, who stands behind it, what does it rest on, and what happens to me if that moves? Nobody is short of data. Everyone is short of the shape of it, so today that question gets answered by hand, out of documents that were never built to connect.

• Investors. What am I actually exposed to? Two holdings can look unrelated and depend on the same supplier, the same lender or the same jurisdiction. The exposure nobody can see is the one that reprices everything at the same time.
• Funds and asset managers. Look-through on demand. Concentration, coverage, mandate fit and jurisdiction mix come from querying the data model instead of from a quarterly reconstruction. The same question, asked at any hour, returns the same answer.
• Public companies. Your own ecosystem is the blind spot. The supplier behind your supplier. The customer who is also a major customer of your competitor. The single dependency that nobody wrote down because no one team could see both ends of it.
• Regulators and supervisors. Systemic risk is a relationship problem. Risk does not concentrate inside one filing; it concentrates between them. A shared data model shows where it gathers, with every claim carrying the source it came from and the date it took effect.
• Banks and lenders. Who am I really lending to? Follow the sureties, the parent, the pledge and any pledge over that pledge. Collateral is only ever as good as the chain behind it, and that chain is a graph, not a document.
• Everyone else in the market. Auditors, exchanges, insurers, counterparties. Anyone whose job is to establish that something is what it claims to be is doing graph work right now, with phone calls and PDFs. The work does not change. The evidence stops being manual.

Not in the rows. In the relationships.

Every fact below was already public. None of it was visible in any single filing, report or dashboard, because none of those hold the shape of the whole. We pointed Round Table at the AI sector using public sources only. These are shapes the graph put in front of the people reading it, none of which had a column anywhere.

• Six household names, one root. Follow the dependencies down and all six names converge on a single firm in Veldhoven that makes the lithography machine every leading-edge AI chip is printed on. It sits behind a Taiwanese fab, behind an American chipmaker, behind the names everyone actually watches. No single filing states the whole chain. The graph lays it out where anyone can see it.

A dashboard answers the question you already knew to ask. A data model you can query also lets you ask the ones you had not thought of.

Open the Live map and check them yourself

One data model across the organisation

The value compounds when this stops being a project inside one department. Point one data model at your customers, your people and everything you depend on, and the questions that used to need a working group and three weeks become a query over the graph.

• Your customers. One customer, not one record per system. Who introduced them, what they hold, who they pay, who pays them, what they are exposed to. Where your own book is concentrated becomes something you can read from the data model rather than a study someone commissions.
• Your people. Authority, not an org chart. Roles, delegation, approval limits and access sit inside the data model. Who can commit the firm to what, on whose behalf, and what they touched. Key-person dependency becomes visible instead of anecdotal.
• Your ecosystem. Including the parts you do not own. Suppliers, partners, distributors, lenders and their dependencies. The concentration you inherit from other people's suppliers is real exposure, and it is missing from most dashboards.

Risk and opportunity are the same question, asked in two directions.

A data model you can query, not a diagram.

A diagram shows a subject. The ontology records it, as a data model of a sector, an institution, a counterparty or a single instrument that you can query.

• The same parts. Entities, not tables. A person, a company, an agent, an asset, an obligation and an authority each exist once, defined once, and are referenced everywhere rather than rebuilt inside every system that needs them.
• The same wiring. Typed relationships. Owns, controls, supplies, backs, is permitted by. Every connection carries its own type and direction, so the structure is something you can traverse instead of something you have to remember.
• The same rules. Constraints in the data model. What has to hold for a claim to stand is part of the data model, not a note in a policy document, so a claim that breaks a rule can be found by checking it against the rules before it becomes a loss.

And the same clock. Every claim in the data model carries a source and the date it took effect, so you can tell what holds now from what held last quarter and from what was never checked at all. A data model of a sector can be built from public data, and one of your own institution from its own records.

Mapping complex instruments

The dangerous instrument was never the complex one; it was the opaque one. Complexity is only composition: claims on claims on assets. An ontology keeps every relationship explicit, so you can see the whole structure and trace it down to the assets it rests on.

Illustrative example. A tokenised fund decomposes into a Senior tranche, a Mezzanine tranche and a real-world asset (RWA) sleeve. The Senior tranche rests on a Reserve (reserve attestation recorded). The Mezzanine rests on a Loan pool (obligors identified). The RWA sleeve rests on Property (custodian and provenance) and a Bond (valuation sourced). Each claim links to its supporting record: the reserve attestation, the identified obligors, the custodian and provenance, and the valuation source.

• Map. Look-through, all the way down. Decompose any instrument into its full dependency graph, every claim, counterparty and piece of collateral, down to the reserve or real asset it rests on. Computed from the graph, not reconstructed by hand in spreadsheets.
• Report. A query, not a reconciliation. Exposure, coverage, concentration, regulatory treatment and jurisdiction mix all come from querying the graph, on demand, instead of a month of chasing the same numbers between teams.
• Understand. Query it, don't just store it. Ask whether the waterfall holds, whether every leg is permitted and whether every claim has its evidence. Contradictions show up instead of staying hidden.

Six changes that came together

None of this was true a few years ago. Together they have outgrown the way most organisations keep their records.

• AI stopped advising and started acting. AI agents can now move money, sign documents and take decisions directly, a new kind of actor that most systems of record have no type for.
• Value became programmable. Money, assets and contracts are software now: a stablecoin, a central-bank currency, a tokenised building, each one a claim that has to be checkable.
• Identity fragmented. A person, a company or an agent is a different, disconnected record in a hundred systems that never agree on who is who.
• Regulation moved closer to the transaction. For many activities, what is permitted, by whom and where has to be checked when the transaction happens, not only in paperwork after the fact.
• Trust stopped scaling. You can no longer meet, or vet, everyone you transact with, let alone every machine acting on their behalf.
• And our records got better at facts than at meaning. Systems became excellent at recording that something happened, and no better at understanding what it was, or how it connects to everything else.

Most systems of record store transactions and balances, not the typed, sourced relationships between them, and relationships are where meaning lives.

Why AI models need a sourced record

A language model reasons without a sourced record to check against. Point it at a business and it will produce confident, well-written answers nobody can act on, because nothing in the system separates a checked fact from a plausible sentence. The ontology gives that reasoning a sourced record to work from. It is the difference between software that can talk about a business and software that can operate inside one.

• Grounded, not guessed. Every input carries a source. An answer built on the ontology traces back to the claim, the source and the date it took effect. Where the evidence runs out, the gap shows up as a gap instead of being written over with something that reads well.
• An AI agent needs structured data. Reasoning alone does not let an agent act. It acts on entities, relationships and permissions it can read and write. Without that layer, every action it takes rests on guesses about a world it cannot see.
• The alternative is not none. A hundred private data models. Every team already carries its own idea of what a customer, an exposure or an obligation is. None of them are written down and no two answer the same question the same way. One shared data model is not extra structure. It replaces the hundred you are already running on.

Keeping control of your knowledge

An AI model is a commodity: someone else's weights, on someone else's hardware, on someone else's release schedule. That is a perfectly good arrangement for the reasoning. It is a terrible one for the thing being reasoned over. Your edge is what you know that nobody else does, and it stops being an edge the moment a third party holds it. KXCO keeps your knowledge as data you hold. An AI model trained on it does not.

• Weights are not a record. Absorbed, not organised. Train an AI model on your business and it has swallowed your knowledge rather than structured it. You cannot cite a weight, correct one fact inside it, or take a customer back out of it. Ask the same question after a version change and the answer can move, with nothing to point at and no way to tell whether you or the AI model was wrong.
• Whoever holds the interface holds the asset. A vendor between you and your own business. If the only way to ask what you own, who you owe and what you are exposed to is a supplier's AI model, that supplier sits in the middle of your own understanding: their pricing, their retention policy, their availability, their roadmap. Keep the knowledge as data you hold and the AI model becomes a component you can replace.
• Evidence, not recall. "The model said so" is not a filing. No supervisor, auditor, board or counterparty accepts a confident summary. They accept a claim with a source and a date attached, and the ability to check it without trusting the party presenting it. That is what an ontology holds. It is not something weights can produce, however fluent they sound.

AI models will come and go. Your understanding of your own business has to outlive them.

So the AI model stays outside the knowledge, and the knowledge stays sourced, dated, portable and readable by whichever AI model you choose to point at it. Swap the reasoning whenever something better arrives. You should never have to migrate your records.

The four roles at the core of the ontology

At the core of the ontology sit four roles: post-quantum cryptography, AI, blockchain and the regulator. In the data model, an issuance is complete only when all four hold together.

• Trust (post-quantum cryptography) secures Issuance.
• Judgment (AI) interprets Issuance.
• Record (blockchain) remembers Issuance.
• Permission (regulation) permits Issuance.
• Issuance is where a token becomes real, only when all four primitives agree at once.
• Herald Verify, from KXCO Herald, is the edge between Judgment and Trust.
• Herald Sign is the edge between Trust and Record, anchoring documents to the record.

• Portcullis Survey, from KXCO Portcullis, connects to Trust.

• Sovereign self-custody, from KXCO Sovereign, is the edge between Record and Trust.

Everything else is an application.

KXCO is more than a blockchain, a wallet or a signing tool, and it is not an AI model. The ontology is the shared data model: the layers above it work from it, and the layers below supply the signatures and records that let its claims be checked. Between judgment and the world sits one governed step, which decides whether a proposal is allowed to become a real, recorded action.

Humans, businesses, governments, machines and AI agents don't get five different views of the world. They work from the same record.

How tokenised assets are represented

Each kind of token carries the same four links: who issued it and under what permission, who holds it, the evidence for its backing, and its recorded lifecycle. Each of these used to be a promise in a PDF.

• Stablecoin. A claim on a reserve: reserve, with the signed reserve attestation recorded; redemption right, on record; issuer, permissioned.
• Central bank digital currency (CBDC). A sovereign liability: central-bank issuer; legal-tender status; jurisdiction and policy controls; signed with post-quantum algorithms intended to stay verifiable for decades.
• RWA token. A claim on a real asset: underlying asset and legal wrapper; named custodian; provenance chain; valuation source.

Operating principles

• The four-agreement rule. In the data model, an issuance is complete only when trust, judgment, record and permission all hold together.
• KXCO is the software layer, never the operator. Licensed institutions hold assets. KXCO supplies the software. The licensed institution holds the licence and the custody; KXCO holds neither.
• Trust is forward-only. Verification begins at go-live. Provenance is never backdated.
• Off-chain proof, on-chain anchor. The chain records the anchor. The proof is checked independently.
• The regulator is a primitive, not an obstacle. Permission is designed in, kept live, cited and current.

How it holds up under scrutiny.

The questions that decide whether a data model is working infrastructure or just a diagram, answered plainly.

• Where it runs. Inside your estate, not ours. KXCO builds the software and licensed institutions deploy and run it, under their own brand, licence and controls. The reasoning points at your own inference endpoint, so the weights stay in your estate too. The argument on this page would be hollow if Round Table ran on somebody else's hardware.
• What happens to your data. It stays data you hold. The ontology is a data model you own, sourced, dated and portable, not a corpus absorbed into somebody's weights. KXCO takes custody of nothing. That is the whole point of keeping the AI model outside the knowledge: you can swap the reasoning and you should never have to migrate your records.
• What you are actually buying. Round Table, the software that builds and keeps the ontology, and the work of building the data model with you, not a subscription to somebody else's answer. It is deployed inside your estate rather than handed over as a download. It can be co-branded, white-labelled or deployed under your own name, the same way the rest of the platform is. The Round Table Live map was built the same way yours would be, from sources you can open. Investment teams can request the thirteen-page pack.
• Where to start. With the question you cannot currently answer about your own business. Bring that, and whatever data you already hold, and the first pass is scoped around answering it rather than around installing something. On public sources, Round Table infers relationships. On your own assets, it works from your own records, so each claim traces to a source you hold.
• How the ontology is constructed. It is a shared schema of the entities that matter economically (identity, authority, assets, events, proof and history) and the relationships between them. Each entity is defined once and referenced everywhere rather than rebuilt inside every system. Where a claim has to be permanent, its proof is signed and anchored to KXCO Armature, the post-quantum record, so anyone with the record and the public key can check it later without asking us.
• How information becomes trusted. Nothing is trusted because KXCO says so. Claims are signed, and where a claim has to be permanent its proof is anchored to Armature. A party with access can then check who signed it and that it has not changed. A signature does not show that the claim is true. Signatures use post-quantum algorithms from NIST FIPS 204 and 205, and key establishment uses FIPS 203, so those signatures are designed to stay verifiable for decades.
• Third-party extension. The ontology is a base others build on, not a closed product. Institutions and developers define their own entity types and relationships against the shared data model, so their systems interoperate by default instead of through one-off integrations.
• Network effects. Within a deployment, each new participant, asset and record makes the shared data model more complete and more useful to everyone already working in it. Because entities reference one another, value grows with the graph: the more of the business the data model covers, the more can be checked and acted upon.
• Portability. Identity, history and proofs accumulate in the ontology and stay independently verifiable and portable; the record is not locked in a vendor silo.

Go deeper in the engineering blog: how AI agents act on the ontology, and why we think BlackRock's quantum warning makes a shared, post-quantum record urgent.

KXCO does not ask you to trust the issuer's claim, or the issuer's reading of the law. It records the evidence for the first, with its source and signatures, so you can check it, and keeps one shared data model, sourced and dated, to reason about the second.

See it on public dataSee the productsBring us your question