Failure Registry Methodology

What is included

A business enters this registry when it held cryptocurrency or money on behalf of customers or creditors, stopped returning it, and owed at least $10 million when it did. Exchanges, lenders, funds, brokers, custodians, miners and token issuers all qualify. A formal insolvency proceeding is not required — several entries here simply closed their doors and never filed anything — but where a proceeding exists, its docket number is recorded so the filings can be pulled directly.

The threshold is higher than the $100,000 floor used in our hack registry, and for a practical reason: a small exploit is still a clean, checkable on-chain event, whereas a small insolvency usually leaves no docket, no schedules and no way to establish what was actually owed. Below this line the long tail would be a list of rumours.

What is excluded. Businesses that wound down while paying everyone back. Startups that failed owing only their investors, with no customer assets involved. Token prices falling, however far — a token going to zero is not a business owing money. Ponzi schemes and investment frauds that never operated a real custody business belong in a different register; where such an operator did hold customer balances and a court appointed a receiver over them, it is in scope.

How claims are measured

The headline figure for each entry is what was owed to customers and creditors, as scheduled by the debtor or allowed by the court. Nothing else.

This rules out three numbers that circulate widely and mean something else entirely. The company's last private valuation is not a liability. The peak market value of the tokens customers had deposited is not a liability. Neither is the “hole” between assets and liabilities, which is a shortfall, not a claim. All three produce larger, more quotable figures than the real one, which is exactly why they need to be named and rejected here.

Where credible sources disagree, the registry publishes the range alongside the figure it considers best supported, and says which source each bound came from. A single number implies a certainty the filings often do not support.

How recovery is reported — read this before quoting a percentage

This is the most misquoted quantity in the subject, and getting it right is the main reason this registry exists.

Most large US estates convert every claim into dollars on the day the case opens, then repay against that frozen figure. If a customer deposited one bitcoin, the claim becomes the dollar price of one bitcoin on the petition date, and it stays there for the years the case runs. A plan that pays 100 per cent — or 120 per cent — of that claim has repaid the petition-date dollars. It has not returned the bitcoin, and if the price rose in the meantime the customer receives a fraction of what they handed over, while the headline reads that creditors were made whole.

Returning the same assets in the same units is a genuinely different outcome, and it is rare. So every percentage in this registry is stored with the basis it was measured against, and the two are never rendered apart — not in the table, not on the entry page, not in the CSV, not in the JSON. Where no source states a basis, the field says so rather than guessing, and the entry is flagged for review.

• Petition-date USD. Claims were fixed in dollars at the date the case opened, and repaid against that fixed figure. Says nothing about the value of the assets originally deposited.
• In kind. Creditors received back the same assets they deposited, in the same units.
• Current USD. Distributions valued at the price on the date they were paid.
• Mixed. Some classes repaid in kind, others in dollars.
• No distributions yet. The estate has not paid creditors anything.
• Basis not established. A figure circulates but no source states what it is measured against.

What counts as verification

An entry is marked verified only with at least two independent sources. Five outlets restating the same filing count as one source, because they are one source. Entries that do not clear the bar remain visible, labelled reported, and are excluded from every published total.

• Court filing. A document on the docket: petition, schedules, disclosure statement, plan, trustee or examiner report. The strongest source this registry has.
• Primary. The estate, the administrator, the company itself, or a regulator: claims portals, official updates, enforcement notices.
• Secondary. Original reporting by a publication that adds information beyond restating someone else.
• Aggregator. A third-party tracker listing the case. Recorded for completeness only — never counted toward verification and never used as a source of figures.

Court filings are the strongest evidence available here and are preferred over reporting wherever both exist. Aggregator trackers are recorded when they help a reader find their way around, but they never count toward verification and no figure is ever taken from one.

Procedures are recorded as they actually are

Insolvency regimes are not interchangeable, and translating them into American terms puts a wrong fact into every citation. A Japanese civil rehabilitation is not a Chapter 11. A Cayman provisional liquidation is not an administration. Each entry names the local procedure.

• Chapter 11. US reorganisation, debtor generally stays in possession of the estate.
• Chapter 7. US liquidation, a trustee sells the estate and distributes proceeds.
• Chapter 15. US recognition of a foreign main proceeding, usually to protect US assets.
• Liquidation. Court-supervised winding up outside the US Bankruptcy Code.
• Provisional liquidation. Interim offshore appointment — common in Cayman and BVI — that freezes the estate while solvency is assessed.
• Administration. UK-style process run by an administrator, aimed at rescue before winding up.
• Civil rehabilitation. Japanese reorganisation procedure. Not equivalent to Chapter 11.
• CCAA. Canada's Companies' Creditors Arrangement Act — court-supervised restructuring.
• Receivership. A receiver is appointed over assets, often at a regulator’s request.
• No formal process. Operations stopped and customers were not repaid, but no insolvency proceeding was ever opened.

How the cause is chosen

Every entry carries one cause: the proximate trigger. Real collapses have several — a hole in the balance sheet, a withdrawal run and absent controls usually arrive together — and attempting to apportion blame across all of them would make the field useless for counting. The registry records what stopped the business paying, and the summary describes the rest.

• Fraud or misappropriation. Customer assets were taken or misused by the operator.
• Hack or theft. A security incident removed assets the business could not replace.
• Contagion. Failure of a counterparty that the business was exposed to.
• Market or leverage losses. Trading, lending or treasury losses from price moves.
• Liquidity run. Solvent on paper, but withdrawals outran liquid assets.
• Mismanagement. Commingling, absent controls or unsustainable economics, without a single trigger.
• Regulatory action. An order, seizure or licence loss stopped the business operating.
• Undisclosed. The business stopped paying and never explained why.

Charges, convictions and named people

Where charges have been brought, the entry states the charge and its current status: charged, pleaded, convicted, acquitted, dismissed. It never asserts guilt before a conviction, and it never characterises anyone's conduct beyond what a filing or a court has established. Every record naming charges is flagged for manual review before publication, and any entry above $1 billion in claims is too.

Corrections

Recovery figures move every time an estate distributes, which makes entries here change more often than in the hack registry. When a published number changes, the entry records what changed, when, and why — and the record of the change stays on the page. Nothing is silently edited.

If something here is wrong, write to [email protected]. Corrections that can be checked against a filing are made quickly, and the change is logged where anyone citing the old figure can find it.

Reuse

The registry is published under CC BY 4.0. Republish it, adapt it, build products on it, including commercially — the only condition is that you credit iTokenly and link back. CSV and JSON exports need no key and no account.

One request rather than a condition: if you chart the recovery column, carry the basis column with it. Without that, the chart is four different quantities plotted on one axis.

What this registry cannot tell you

It is not complete and does not claim to be. Failures below the threshold are out of scope by design. Failures in jurisdictions that do not publish filings are under-represented, because there is nothing to check against. Businesses that quietly stopped paying without any announcement may never appear at all.

It also cannot tell you what an individual creditor will receive. Claim classes differ, preferences and clawbacks vary by regime, and an estate-wide percentage is an average across creditors whose positions are not alike. For an actual claim, the estate's own claims agent is the authority; this registry is a reference, not advice.