Machine-Understandable Markets

On November 9, 2026, the SEC's Division of Economic and Risk Analysis holds a Roundtable on Machine-Understandable Data in the Age of Super Intelligence. The question it poses is how market participants use machine-understandable data and super intelligence, and what it costs them.

There's a more fundamental question underneath that discussion: how much of the financial data we produce today can machines reliably understand?

For much of the market, the answer is still surprisingly little. Models are becoming increasingly capable, but the financial infrastructure underneath them was built for a different kind of reader.

Financial data was built for people, mostly for lawyers

For most of its history, financial disclosure has been written to be read by a person and defended by a lawyer. Filings are prose. Machine-readability was added later, as a layer on top.

XBRL made financial disclosures significantly more accessible to software, but also exposed the difficulty of standardizing information that was never designed around a common data model. Tagging introduced new costs, particularly for smaller filers, while custom extensions allowed companies to define their own metrics and tags, making comparisons harder than they should be.

Extracting numbers became easier. Understanding whether they mean the same thing across companies still requires additional work.

Unreadable data is expensive, and it favors incumbents

Reading the market today takes compliance departments, specialized filing services, and paid data subscriptions. Large firms can carry that overhead. Small firms cannot.

The same asymmetry shows up in how standards get set. Financial identifiers are often proprietary and paid for, and the incumbents who sell them have every reason to protect that franchise. A market's basic reference data should not be a toll road.

Onchain finance offers a different starting point. The underlying ledger is public, and anyone can inspect the same transaction history. Processing that data at scale still requires infrastructure, but access to the source record doesn't depend on a proprietary subscription.

The reader is now a model

For most of market history, financial data was consumed by human analysts reading at human speed. Today, models are increasingly used to extract information, reconcile records, monitor activity, and support financial decisions.

A model can process millions of records, but that doesn't help much if the same asset has different identifiers across systems, or if two datasets use the same metric to describe different economic activities.

Better models alone won't resolve inconsistencies in the data underneath them.

Onchain finance is already born structured

A blockchain is not a document you parse. Transactions, logs, and state changes are recorded in structured, programmatically accessible formats, with timestamps and cryptographic evidence that allows the underlying activity to be independently verified.

Unlike a financial filing that needs to be converted into machine-readable information after publication, the blockchain record is structured from the moment it is created.

Onchain finance gives us a working example of financial infrastructure where structured data is part of the system itself.

But structured is not the same as legible

Raw onchain data is structured and still close to unusable. There are more than 150 chains, each with its own quirks. Addresses have no names. Events have no prices. And a blockchain records what happened technically, which isn't always the same as understanding what happened economically.

Stablecoins offer a good example. In our latest State of Stablecoins & Payments research, we found that $85 trillion in reported transfer volume translated into roughly $4 trillion in real economic activity after filtering out non-economic transfers. The blockchain recorded those transactions correctly, but understanding what they represented required consistent definitions and classification.

The same challenge appears across onchain markets. One asset can exist across multiple chains and wrappers, each with different contract addresses. Making sense of that activity requires normalization, shared identifiers, and enrichment that turns raw events into priced, attributed facts.

This is the infrastructure institutions and regulators need to work with onchain data at scale, and it's what we build at Allium.

Designing data for its reader

We call this discipline data UX: shaping the data for the person or model consuming it, so they have to do as little work as possible.

One of our standing rules is that a query should never require one more join than it needs. For an analyst, that means less time reconciling datasets. For a model, it means consistent definitions, reliable identifiers, and enough context to interpret the results.

The next financial standard will be built for machines

As AI takes on a larger role in financial markets, making more data available won't be enough. Markets need shared identifiers, consistent definitions, and infrastructure that allows different systems to interpret the same financial activity reliably.

Standards like FIGI have already shown why common identifiers matter in traditional markets. As financial activity moves onchain, we need that same consistency across blockchains, tokenized assets, and the institutions interacting with them.

That's where the next layer of financial infrastructure needs to be built. At Allium, we're working to connect onchain activity to the identifiers and definitions institutions already rely on, so that assets, transactions, and market data can be understood consistently across systems.

If machines are going to play a larger role in financial markets, they need a common language for what they're reading. Building that language is the work ahead.