Digital Asset Custodian: What Segregation Means When It Matters

Segregation is the word that does the most work in a custody agreement and the one most often accepted without examination. It decides what you own when the custodian stops operating. It is worth pinning this to something concrete, and a licensed digital asset custodian publishes the numbers the argument turns on.

The three arrangements

Commingled. Client assets and the custodian’s own assets in the same wallets, with internal records distinguishing them. If the custodian fails, there is a real argument about which assets are whose, and clients become creditors in a process that takes years.

This arrangement should disqualify a provider for any meaningful amount.

Omnibus segregated. Client assets pooled together in wallets identifiably held on behalf of clients, entirely separate from the custodian’s own assets. Internal records attribute holdings to each client.

This is the standard arrangement and it is sound, provided two conditions hold: the wallets genuinely contain only client assets, and the attribution records are accurate and independently verified. If a shortfall arises, clients share it proportionally rather than by who withdrew first.

Individually segregated. Wallets attributable to a single client. Cleanest position, highest cost, typically available above a size threshold.

How attribution actually works

In an omnibus arrangement the blockchain shows a pooled balance. Your claim to a share of it lives in the custodian’s records. Merchants meet this from the other direction, and a platform set up for client account handling is the usual answer there.

This is the same model as securities custody in traditional finance and it works for the same reason: the records are audited, the reconciliation between internal records and on-chain balances is performed frequently, and a regulator can inspect both.

The question to ask is how often that reconciliation runs and who checks it. Daily automated reconciliation with an independent assurance report over the control is what good looks like. Monthly manual reconciliation is not.

What happens in insolvency

Under a framework with proper client asset rules, segregated client assets do not form part of the insolvent estate. An administrator identifies client assets, verifies entitlements against the records, and returns them. Clients are not competing with the custodian’s creditors.

The process still takes time. Months, not days, because entitlements have to be verified before anything is returned. But the outcome is return of assets rather than a distribution of cents.

Where this breaks down is a shortfall: if the pooled wallets hold less than the records say clients are entitled to, the shortfall is shared. That is why the reconciliation frequency and the assurance over it matter so much. They are the controls that prevent a shortfall existing undetected.

What to confirm in writing

Which arrangement applies to your account specifically, because a provider may offer several. Where the assets are held legally and under which jurisdiction’s client asset rules. How often reconciliation is performed and who provides assurance over it. What the position is on a shortfall. And whether the custodian is permitted to use client assets for any purpose, which should be a clear and unqualified no.

That last point deserves emphasis. Rehypothecation, using client assets to support other activity, is how several failures became insolvencies. A custody agreement should prohibit it explicitly, and you should read the clause rather than accept a verbal assurance.

The practical summary

Segregation is not a feature, it is the product. A digital asset custodian operating under a European client asset framework is selling you a legal position, and the technical security is what protects that position rather than what constitutes it.

When comparing custodians, compare the legal arrangement first and the technology second. The technology differences between serious providers are small. The legal differences are not.