Crypto Platform Failure: Recovery in 2026

Introduction

Search for what creditors of a collapsed crypto lender got back and the results are confident: thirty cents on the dollar, or sixty, or a figure carried to one decimal place. They are specific, they are widely repeated, and almost none of them can be traced to a document that says what they claim. That is not a gap in the coverage. It is a property of the subject, and it is what this article is about.

The reason is structural rather than conspiratorial. An insolvency estate does not publish a single recovery number, because for most of its life it does not have one. It distributes in rounds, over years, against a claim pool that keeps changing as claims are allowed, disallowed and forfeited, and it pays in whatever it has: dollars, cryptocurrency, shares in a successor company. Its progress reports are written in careful language that summaries flatten.

On 21 August 2026, this project's own research pass read 358 primary documents — court filings, trustee notices and claims-agent pages, across the estates named on this page — and did not return one defensible recovery percentage. That count is the one number here that does not come from a court filing, and it is stated with its denominator and its date for the same reason every other figure is.

What came back instead was better material: petition dates, claim windows with hard deadlines, the form each distribution took, the size of the population that never collected, and a small number of percentages that do survive checking, set out below with every bound their sources attach.

Our guide to lending risks and insurance sets out the four distinctions that decide whether anything is backed before a failure, and our broader lending risk guide covers what can go wrong in a position before any estate exists. This article takes up the story afterwards, when the only remaining question is what a claim converts into and when.

Three constraints govern what follows. Every distribution figure comes from a court filing, a trustee statement or a court-appointed claims agent, with a date. No news article, aggregator or summary is used at any point. One source below is none of those three: a Federal Reserve bank letter. It is admitted because it rebuilds the collapses from the debtors' own bankruptcy schedules, and because it marks which of its numbers are its authors' estimates. Those are the qualifiers a summary drops first.

No platform is recommended, and several named here are defunct, in liquidation or in wind-down. Where a figure could not be traced, it is described in words rather than guessed at. Most platforms below held customer assets in their own custody, which is why a depositor became a creditor rather than a person withdrawing a wallet balance. Two named here did not. Terraform Labs runs a claims process for holders whose coins were locked on its own network rather than deposited with it. Three Arrows Capital was a fund with no retail depositors at all.

What 358 primary documents returned

The research pass of 21 August 2026 covered the estates of the 2022 collapse and those that followed. Its brief was narrow: for each estate, find a recovery percentage stated in a primary document, with the base it was computed on and the date it applied to. Across 358 documents that combination did not come back cleanly once. What the pass produced was dates, deadlines, payment mechanics and disclaimers, and a shortlist of percentages worth attacking properly. The three figures below came from two narrower checks that followed. An adversarial re-examination running to 150 primary-source fetches closed on 21 August 2026, and cleared two of them. A reading of a claims agent's case site on 23 August 2026 produced the third.

The disclaimers are not evasions. An estate that reserves its rights over a cumulative recovery calculation is saying the calculation is unstable, which is more useful than a number would have been, and the figures that circulate are frequently that arithmetic with the warning removed.

An institution with more resources hit the same wall

This is not a limitation of one research pass. The Federal Reserve Bank of Chicago rebuilt the 2022 runs directly from the bankruptcy filings, in Chicago Fed Letter No. 479 of May 2023 by Radhika Patel and Jonathan Rose. The letter reconstructs the sequence in detail: which platform paused withdrawals when, how much left each one after the TerraUSD depeg and after the failure of Three Arrows Capital, and how many customers each estate owed money to at the moment of filing.

It publishes no recovery figure at all. Its closing description of the people involved is that they are "waiting for the return of whatever can be salvaged of their investments as bankruptcy proceedings conclude". Two authors with access to the full filings, writing for a Federal Reserve bank, characterised the outcome in exactly those terms rather than in cents on the dollar.

The population counts, which are checkable

What the same letter does supply is the size of each creditor body, taken from the debtors' own Official Form 206 schedules. These are counts of customers with positive claims at the moment of filing, not of everyone who ever held an account, and the letter says so explicitly. Read on 23 August 2026, its figure 1 records:

• Celsius — withdrawals paused 12 June 2022, chapter 11 petition 13 July 2022, 542,333 customers owed debts in the filing.
• Voyager Digital — withdrawals paused 1 July 2022, petition 5 July 2022, 975,521 customers owed debts in the filing.
• BlockFi — withdrawals paused 11 November 2022, petition 28 November 2022, 599,766 customers owed debts in the filing.
• Genesis — withdrawals paused 16 November 2022, petition 19 January 2023, approximately 340,600 customers, most of them reached through a partner platform rather than directly.
• FTX — withdrawals paused between 8 and 11 November 2022, petition 11 November 2022, 1,869,269 customers owed debts in the filing.

Those five numbers behave the way a figure should. Each has a stated source document, a defined population and a fixed date, and any reader can open the letter and disagree with our reading of it. Not one of them is a recovery rate, and the letter never implies otherwise.

What the same letter says about the runs

The letter also sets out the structural reason these estates exist, and it is a liquidity story rather than a fraud story. The platforms let customers withdraw on demand while placing the same money in illiquid and risky investments, in part to fund the yields advertised on the deposit products. That is a classic maturity and liquidity mismatch, carried out on assets far more volatile than the ones the pattern is usually described with.

Its figures for how the money left are drawn from the same filings. Celsius had placed about 935 million dollars in TerraUSD and the Anchor Protocol, which offered returns of up to 20 per cent on that stablecoin, and on the letter's own estimate saw about a fifth of customer funds withdrawn in the eleven days after the depeg of May 2022. BlockFi reported outflows of 4.4 billion dollars between January and May 2022, before its November bankruptcy. In the sharpest single episode, customers withdrew a quarter of their investments from FTX in one day, on the letter's analysis of that estate's own bankruptcy filing.

None of that is a recovery figure either, and it matters for reading the ones below. A recovery percentage here depends on what illiquid assets were eventually sold for, years later, by somebody else. It is an outcome of a liquidation, not a property of the deposit.

What counts as a primary document here

Insolvency estates publish in a small number of standard formats, and knowing which one a claim came from settles most arguments about it. Five recur here, each answering a different question.

• A status report to creditors is the Plan Administrator's periodic account of what has happened. It is where combined totals, cheque counts and outstanding balances appear, written in the past tense about events that have occurred.
• A notice of distribution announces a round. Its tense is the thing to read: a notice may say the debtors intend to commence, which is a statement about the future and not evidence of a payment.
• A post-confirmation report is a quarterly financial return recording cash disbursed in the quarter and cumulatively, so it can corroborate or contradict a claim that a large round happened in a given period.
• A bar date order fixes the deadline for filing a claim and states what happens to anyone who misses it. It decides whether an entitlement exists at all.
• A court-appointed claims agent's case site republishes the above for creditors, and adds the operational detail — payment rails, redemption windows, support routes — that rarely reaches the docket in a readable form.

None of these is a press release, and none is trying to summarise the case for a general reader. That is why they are usable: the qualifiers are still attached, because the audience is people whose money is in the estate.

Why secondary sources are excluded entirely here

A summary of an insolvency drops the qualifiers first, because they are the least quotable part of the document: the entity, the cohort, the base, the tense. What survives the compression is the number, which is the part that means least on its own.

Four ways a recovery figure stops meaning what you think

Recovery does not mean what most readers assume, and it stops meaning it in four distinct places. Each has named instances in the filings, and each can turn a headline percentage into a poor description of an individual outcome. Together they are the reason a single number cannot answer the question a depositor is actually asking.

1. The claim is fixed in dollars on the filing date

In a United States chapter 11 case a claim is valued in dollars as at the petition date. For Celsius that date is 13 July 2022 and for Voyager it is 5 July 2022. Everything the estate reports afterwards is a percentage of that frozen dollar figure, whatever has happened to asset prices since.

The consequence is that two true statements describe one outcome. A claimant paid the whole of a petition-date claim has recovered in full, and has also lost every unit of the exposure they originally held. Estate reporting uses the first framing because it is the one the court process is denominated in. A depositor who thinks in coins rather than dollars is reading an answer to a different question from the one they asked, and no footnote in the filing will tell them so.

The choice of unit survives into the payment. BlockFi's wind-down distributions are made in United States or Canadian dollars by claimant location, on the claims agent's own account, and offer no route back to the original cryptocurrency.

2. Coverage can exist and still be unreachable

The second mechanism is the one that costs retail claimants the most, because it is invisible until it has already happened. An entitlement is real, the money exists, and a deadline passes. Three instances in the record show three different shapes of the same problem.

The cleanest is the Terraform Labs wind-down trust, whose crypto-loss claims bar date fell on 16 May 2025 at 11:59 p.m. Eastern time. The court-appointed claims and noticing agent states the consequence in the order's own words: if proofs of claim are not received on or before that date, "the holders of the underlying claims will be barred from asserting such claims against the Debtors and receiving distributions from the Wind Down Trust on account of such claims in these chapter 11 cases". Relief afterwards requires a motion, which the Plan Administrator may oppose. One date, one agent, one stated consequence.

Celsius is harder, and worse for the claimant, because its window is not a single public date. Under forfeiture procedures the court approved on 12 February 2025, two fixed deadlines applied to creditors whose distribution was first attempted on or before 29 February 2024: 31 March 2025 to redeem it or open a support ticket, and 30 June 2025 to answer any follow-up request. Every other cohort runs on a private clock — "at least one year after the e-mail from Celsius", or "at least 90 days after the e-mail is sent" where no distribution could be attempted at all. A deadline nobody can look up is a harder problem than one they missed.

Behind all three sits the earliest clock, and the one most easily missed because it runs before any distribution is discussed. The docket for the Voyager chapter 11 cases records a last date to file claims of 3 October 2022, roughly three months after the petition. Everything else described on this page applies only to people who were inside that date. A depositor who assumed the estate held a list of account holders and would simply pay them was, in the ordinary case, expected to file.

The BlockFi estate produced the third shape, and it is the one this page cannot pin to a day. An identity-verification deadline passed in May 2025. No day is printed here for two reasons that compound each other: the estate's own announcement names one date in its body and a different one in the web address the publisher gave it, and the only copy of that announcement this page could read is a syndication mirror, which is a secondary source and barred at every point on this page. The claims agent's case site carries no identity-verification material at all. A claimant trying to comply could not have determined from the publisher which clock bound them.

3. The form of payment, and the date it was valued at

Recovery arrives as cash, as cryptocurrency in kind, as equity in a successor company, or as a mixture, and the form changes what the percentage is worth. This is the column that summaries drop first, and it belongs beside every figure rather than in a footnote.

Celsius is the sharpest illustration. Its plan distributed liquid cryptocurrency alongside shares in a successor mining entity, and the plan ascribed a fixed dollar value to that stock. This page carried the figure and its docket number until 23 August 2026, when neither could be retrieved: the filing returned 404 from the free docket mirror and the search surface refused automated access. The figure is therefore not printed, on the same rule that removed six others. What survives the check is the shape, and it is the part that matters: no public market existed to test any ascribed price until the successor listed on Nasdaq in July 2026, so for roughly two and a half years a number was attached to the equity leg and a price was not.

Cohort matters as much as form, and the BlockFi wind-down splits its claimants on the line that mattered most at the platform: whether the assets sat in a custody-style wallet account or in the interest-bearing product. The claims agent's case site puts wallet clients who did not withdraw before the wallet withdrawal window closed in a separate queue from the interest-account claimants who receive the estate distributions. Two people who thought of themselves as customers of one company have different entitlements because of which product held their cryptocurrency.

Payment mechanics matter as much as the asset. A BlockFi claimant who does not select a payment method inside the window is paid by one of two default rails, depending on whether they are in the United States, and the claims agent's page adds that unclaimed funds on the domestic rail revert to the estates 14 days after issuance. A distribution can therefore be issued, defaulted to a rail nobody monitors, and returned, with no deadline in the plan having been missed.

4. Distributed is not the same as received

The fourth mechanism is the one no summary reports, and it is large enough to change what a headline percentage means. The Voyager estate's Eighth Status Report to Creditors sets it out in figures: approximately 495,000 cheques were mailed for the second distribution, for a total of about 587 million dollars, and as at 23 September 2024 approximately 148,000 of them, totalling roughly 83 million dollars, had not been deposited. That is about 30 per cent of the cheques, and the same report puts the deposited total at about 504 million dollars — roughly 86 per cent of the round by value. Sixteen thousand had come back from the postal service as undeliverable, and almost 90 per cent of the outstanding cheques were for less than 1,000 dollars.

That gap was still open sixteen months later. The Plan Administrator's Tenth Status Report to Creditors, filed 27 February 2026, records approximately 65,000 creditors with distributions totalling over 11 million dollars not yet deposited as at 10 February 2026, and set 31 March 2026 as the cancellation date for outstanding second-distribution cheques. The money had been calculated, approved, printed and posted, and had still not arrived.

A distribution can also be half-collected. The Celsius claims administrator notes that a claim distribution is sometimes processed as more than one transaction — a creditor paid through a consumer payments application might receive one amount in bitcoin and a second in ether — and that redeeming only the first leaves the second exposed to forfeiture. A claimant who saw money arrive and assumed the matter closed is exactly who the forfeiture procedures catch.

Celsius handles the same problem by taking the money back. Unclaimed distributions there are not held indefinitely: under the procedures approved on 12 February 2025, forfeited funds are, in the claims administrator's own words, "redistributed to other creditors who have valid claims or used to fund Wind Down Expenses". Both halves of that disjunction are live, and stating only the first turns an either-or into a promise the document does not make.

The figures that survived a primary-source check

Three figures below carry a base, a date, a form of payment and a primary document, and each is written with the bounds its source attaches. Three rows is what survived the checks described above, and the table is deliberately not padded to look fuller. A row a reader can trace is worth more than four they cannot.

Voyager creditors: the total, and its three limits

The combined figure is not our arithmetic. The Plan Administrator's Eighth Status Report to Creditors, filed 25 September 2024 in the chapter 11 case of Voyager Digital Holdings, No. 22-10943 in the Bankruptcy Court for the Southern District of New York, states it directly. Combining the two distributions, creditors of Voyager Digital, LLC who received the second distribution had received a total of 70 per cent of their allowed claim as at September 2024 — a running total, not a final recovery.

Every clause in that sentence narrows it, and a shorter version would be wrong in the opposite direction. The report's own phrase is "distributions to date". The entity named is Voyager Digital, LLC, not every company in the group. And the population is the creditors who received the second distribution, which excludes the 148,000 uncashed cheques described above. A reader told that Voyager creditors recovered 70 per cent will hear none of that.

The supporting sequence is on the docket. Withdrawals were paused on 1 July 2022 and the petition followed on 5 July 2022. An initial distribution of 35.72 per cent was made under Doc 1459 of 14 June 2023, and the Voyager app reopened for in-kind crypto withdrawals on 23 June 2023 for a 30-day period ending 23 July 2023 under Doc 1480. The second distribution began on 31 July 2024 at 34.28 per cent, totalling approximately 589 million dollars, which is the round the Eighth Status Report was written about.

The same report shows why a running total stays a running total. Recoveries from the Three Arrows Capital insolvency, where the wind-down debtor sits on the creditors' committee, were to be passed on to Voyager creditors in later distributions. On the outlook, the Plan Administrator wrote that it "would be premature at this time to speculate on what impact the aforementioned matters might have on creditor recoveries" — a person with every document declining to state the number that circulates without one.

One caution travels with the figure. The docket's own header recorded 10 August 2026 as the date of the last known filing when this page was written on 23 August 2026, and no eleventh status report and no third-distribution notice appear on it. Recent entries are not all available as free documents, so a search that returns nothing is weaker evidence than a filing that says something. The 70 per cent is stated here with its date attached and never as a standing present-tense fact.

Celsius creditors: notice dates against payment dates

Celsius reports cumulative recovery percentages on petition-date claim value as at 13 July 2022, counting cash and liquid cryptocurrency only. Two such levels can be traced to a document here. A cumulative level of approximately 60.4 per cent, roughly 127 million dollars, was noticed in Doc 7871 on 27 November 2024. A cumulative level of approximately 64.9 per cent, 220.6 million dollars, is carried by Doc 8188 of 19 August 2025, a notice of commencement of the third distribution. Doc 8314 refers back to that round as accomplished. An initial distribution preceded both. No docket number for its percentage came back from the research pass, so it is not printed here.

Notice dates and payment dates are two different things, and conflating them is how a projection becomes a recovery. Doc 7871 said the debtors intended to commence the distribution in the near term. Commencement is evidenced by the later filing: Doc 8188 records that the Plan Administrator commenced the second distribution on or around 6 December 2024. Nine days separate the announcement from the event in that instance, and elsewhere the gap has been longer or has never closed at all.

The estate's own concession on those percentages is better evidenced than the percentages themselves, and it is the single most useful sentence in the whole set. Docs 8188 and 8314 both carry it, in identical language:

Because Claims that were forfeited were included in prior cumulative recovery calculations, and due to the variable nature of Liquid Cryptocurrency prices, the Post-Effective Date Debtors had to make certain adjustments when calculating a cumulative recovery percentage... First, the Claims denominator... was reduced by, among other things, the amount of any Class 2, Class 5, Class 7, Class 8, and Class 9 Claims that were forfeited... Second, the numerator... was increased by (i) the current market value of the assets held for Convenience Claims that were forfeited...

The passage continues into a second limb of the numerator adjustment, adding the gain on those same assets as liquid cryptocurrency prices moved, and both notices then reserve all rights in connection with the calculation of the cumulative recovery percentage. Read together, those passages say that a rising sequence of cumulative percentages is not a straightforward measure of creditor progress: the denominator shrinks as claims are forfeited, and the numerator moves with cryptocurrency prices. Some of the rise is arithmetic rather than money.

A fourth round was announced in January 2026 for commencement the following month, and its notice was titled differently from the two before it, which had each announced a commencement rather than a round. The percentage it carries is dealt with below and is printed nowhere on this page, in any tense.

Checking any of this without a subscription

Every document cited here was reached without paying for court access. Dockets for United States bankruptcy cases are mirrored publicly with their entry lists intact, so a case number and a document number are enough to confirm that a filing exists and when it was entered. Reading the document itself is a separate question, because coverage thins for recent entries: on the Voyager docket the Eighth Status Report is readable in full while several later ones are listed but locked. That is why a search returning nothing is weaker evidence than a document that says something.

BlockFi wind-down: a cohort figure with a cap

The BlockFi wind-down produced the third traceable figure, and it is unusual in being defined by cohort rather than by round. Claimants placed in or opting into the Convenience Class are entitled to a 50 per cent distribution on allowed claims, up to 1,500 dollars, paid in United States or Canadian dollars by location. That is stated by the court-appointed claims agent on the case distributions site, which also records that distributions to that class commenced in February 2024 and are continuing, read on 23 August 2026.

A cap changes the meaning of the percentage completely, and it is exactly the sort of clause a summary drops. For a claim of 2,000 dollars the cohort figure is 50 per cent; for a claim of 20,000 dollars the same rule pays 7.5 per cent. Neither claimant has been treated inconsistently, and any average across them would describe nobody.

Estates still distributing: a timeline, not a percentage

Several of the estates most often quoted have no final figure because they are not finished. For these, a date and a status are honest and a percentage is not, so the record below is written as a timeline, and none of it is an estimate of what the outcome will be. Unlike the figures above, none of these lines carries a filing number. Each is what the research pass of 21 August 2026 recorded from trustee and estate surfaces.

• Mt. Gox — the trustee is still distributing more than a decade after the 2014 bankruptcy, against a repayment deadline of 31 October 2026 that has already been extended more than once. It is the estate people reach for when they want a number for how long recovery takes, and it is the worst possible source for one: its end date has never been fixed.
• FTX — the wind-down trust ran a fifth distribution on 31 July 2026. A cumulative percentage for this estate did not come back in a form that carries a base and a date together.
• Celsius — a fourth distribution was announced in January 2026 for commencement the following month, and nothing on the docket confirms that it commenced.
• Vauld — a second distribution is in progress under the Singapore scheme of arrangement approved in 2023.
• Three Arrows Capital — the liquidation feeds other estates rather than retail claimants directly. Recoveries there were to be passed to Voyager creditors in later distributions, which is one reason the Voyager total was never presented as final.

The last line explains a structural feature of these numbers. Estates are chained: one wind-down is a creditor of another, so a figure published by the first cannot be final while the second is still litigating. Reading any of these percentages as an outcome requires knowing that no upstream estate still owes it money, and that condition is rarely satisfied when the number is quoted.

Five ways a recovery figure goes wrong

Six recovery figures used to sit on this site's own guide to lending risks and insurance, linked above. All six were removed in August 2026 because none could be traced to a primary source that said what the sentence around it claimed. They are worth publishing as worked examples, because between them they show five distinct ways a recovery number goes wrong, and every one of the faults is visible in the arithmetic without any specialist knowledge.

A figure can be wrong by being early

The first said that Voyager account holders eventually recovered 35 to 40 cents on the dollar. Nothing about it was invented. It was true, once, of the initial distribution of 35.72 per cent, and it went false the moment the second landed at 34.28 per cent in July 2024. Written with "eventually" in front of it, a first-round figure became a claim about the end of a process still running, and it understated the outcome by roughly half — in the direction that makes a failed platform look worse than the record supports.

Two rows of one table become a range

The second quoted Celsius creditors as recovering "roughly 57 to 73 per cent". Both endpoints exist and neither is an endpoint. They are two rows of the same illustrative waterfall in the estate's effective-date notice: the lower is the liquid cryptocurrency leg on its own, the higher that leg plus a successor-equity leg valued at a stipulated share price. Adding a row to a subtotal does not produce a range. It produces one larger number describing a different thing, and quoting the two together implies a spread of creditor outcomes the document never asserted.

An invented range that contains the truth

The third said Celsius creditors received 50 to 70 cents on the dollar after eighteen months. No primary filing states 50 per cent, or 70 per cent, or any range at that point in the case. What the record produces is single dated levels, such as the approximately 60.4 per cent cumulative noticed in Doc 7871 on 27 November 2024. That figure sits comfortably inside the invented range. It is neither endpoint, and it belongs to a different moment. That is exactly how a fabricated range survives review: a spot-check finds a real figure somewhere inside it and reads that as confirmation. A range that contains a true number is not thereby a true range.

A number bolted to a date it had not reached

The fourth attached "approximately 60 to 73 cents on the dollar" to January 2024. The nearest real figure to the low end, approximately 60.4 per cent, was not noticed until 27 November 2024 and not commenced until on or around 6 December 2024 — eleven months after the date the sentence attached it to. No filing has ever stated 73 per cent. A dated wrong figure is worse than an undated one: the date is what makes the sentence feel sourced, and it invites a check that returns a contradiction.

A worked example in the wrong currency

The fifth turned the fourth into arithmetic: a 50,000 pound balance producing an eventual recovery of roughly 30,000 to 36,500 pounds. It inherits the error in its inputs, and it fails again on its own terms even if they had been sound. A claim fixed in dollars at the petition date and paid out over years in cryptocurrency, cash and successor equity leaves a British claimant's sterling outcome depending on the exchange rate on the day each tranche landed. There is no single sterling number to state, and stating one invites planning against a precision the process does not have.

A generalisation with no population

The sixth said failed platforms historically recover 30 to 60 cents on the dollar. Historically over which cases, across what date range, weighted how? None of that was defined, and a figure with no stated "of what" cannot be checked and therefore cannot be shown to be wrong. That is why it survives and spreads, and wide repetition is the property unverifiable numbers have the most of.

One pattern connects all six, and it is not carelessness. Each reads as more informative than the truth it displaced. A range sounds like a considered estimate where a single number sounds like a coincidence; a date sounds like a citation; a worked example in the reader's own currency sounds like generosity. They are improvements in tone at the cost of accuracy, which is why they survive editing by people who are paying attention. The defence is the specific habit of asking, every time, what the percentage is a percentage of.

The seventh figure, which is not printed here

One more circulating percentage was examined and refused. A Celsius notice of January 2026 carries a cumulative recovery percentage for a fourth distribution that no later filing confirms has commenced. It is a projection, and a projection written as a recovery is the fourth failure mode above wearing different clothes: a figure bolted not to a wrong date but to an event nobody has shown happened. Deleting six figures for that fault and then transcribing a seventh with the same fault would have been the most expensive sentence here, so the number appears nowhere: not on this page, not in the guide that fed it, not in a footnote.

Protection claims that outlived the platform

Nothing takes a marketing claim down when a platform stops operating. Two of the lenders named in this article still serve full commercial websites advertising the protections they sold. One is in liquidation. The other has published nothing on its own site since 2023. Both were read on 23 August 2026 and the quotations below are verbatim from the live pages. This section reports what those sites currently say; it is not a description of what either company can currently do.

An insurance figure, and the two documents behind it

The CoinLoan site describes a service in the present tense, three years after the newest dated item in its own media centre. It says the company "stores clients' assets securely with $250M insurance", advertises a "Certified custodian with $250 million insurance", and calls itself "a regulated financial institution". The badge strip under the headline asserts a European financial licence, a FinCEN money services business registration, and status as an insured digital-asset custodian. Interest accounts at 9.2% p.a. and crypto-backed loans at 5.5% APR are still advertised on the front of it. The copyright line in the footer ends in 2023.

The most useful part is checkable without adjudicating whether the insurance ever existed. The site's own licences page states that "these documents issued by authorized jurisdictions authenticate CoinLoan's legal compliance", and then lists exactly two. The first is a FinCEN money services business registration, dated 2021, which states on its own face that inclusion on the registrant search page "is not a recommendation, certification of legitimacy, or endorsement of the business by any government agency", and adds that FinCEN "does not verify information submitted by the MSB". The second is a certificate of registration for a European Union trade mark, issued by the intellectual property office in Alicante.

So the only European document behind the European financial licence badge is an intellectual-property registration, and the only regulator-issued one says on its face that it endorses nothing. Neither is a finding about insurance. They are a finding about how far a badge can travel from the document under it, and a reader can confirm both in about two minutes.

One thing we could not establish, and worth stating rather than glossing: the logged-out application surface still renders. Its login and registration forms load and respond, so a claim that the software is dead is not something this article can support. We hold no account and created none, so what happens after authentication is untested here.

A page that contradicts itself in two screens

Hodlnaut, in liquidation in Singapore, presents the opposite failure. The halt is not concealed at all: the first three pieces of text in the body of its home page are a bracketed importance marker, the sentence "Withdrawals, token swaps and deposits are halted", and a link to read more. The footer names the company in liquidation, care of a restructuring firm in Singapore. That is disclosure, and more than the other site offers.

The defect here is internal contradiction rather than concealment. Hodlnaut is in liquidation, and its home page still sells the product. On the same page as that banner, its browser tab and its headline advertise earning up to 7.25% APY on deposited cryptocurrency. Its statistics strip claims 250 million dollars in held assets and more than 10,000 active users. Its feature list includes the line "Deposit and withdraw anytime", and, on the same list, "Optional Nexus Mutual's Custody Cover" — a platform in liquidation still listing purchasable cover. One page, at one moment, telling a visitor both that deposits and withdrawals are halted and that they can deposit and withdraw at any time. Its sign-up route no longer works either. Opened in a browser on 23 August 2026, it renders a page headed "Service Unavailable". The site's own FAQ still directs a reader there.

Both sites share a structural feature worth naming. The claims that remain visible are the ones a company controls itself, and the corrections are held by parties with no reason to update someone else's marketing: a liquidator, a court, a claims agent. Nothing links the two. A domain renewal is cheap and a hosting bill is small, so a site can outlive the company by years without anything appearing to be wrong.

What this means for a reader who is choosing today

The point is not that these two sites are unusual, but that a depositor cannot distinguish a real guarantee from an advertised one by reading the advertiser, in either direction. One advertises protections its own supporting documents do not support; the other discloses a halt and contradicts it a screen later. What is resolvable is the document under the claim: every check above was a matter of clicking through to the artefact and reading what it says about itself. Our exchange security checklist turns that into a set of things to verify before depositing. The parallel list of errors people make on the way in is set out in our guide to the mistakes that cost people money.

Conclusion

Three hundred and fifty-eight primary documents produced no defensible recovery percentage, and two much narrower checks afterwards produced three. That ratio is the finding rather than a disappointing result. The record is rich in exactly the material that decides an individual outcome — petition dates, cohorts, deadlines, payment rails, forfeiture rules — and thin in the single summary number that every article about a collapse leads with.

So the useful test on any recovery figure is not whether it sounds plausible. It is whether the sentence carrying it names four things: the base it is a percentage of, the date it was true on, the form the payment took, and the document that states it. A number missing any of those is not a small approximation of a good one. It is a different kind of statement, and the six deleted from our own material were each wrong in a way that a reader could have caught from the arithmetic alone.

The second habit is tense. Estates announce and then, sometimes, commence. A notice of intention is not evidence of a payment, and the gap between them has run from nine days to never. Before repeating a recovery-shaped number, find the filing that says the money moved rather than the one that says it was going to.

And the last one, which costs nothing: distributed is not received. As at 10 February 2026, tens of thousands of people in one estate alone had been sent money that was calculated, approved and posted to them, and had not deposited it. Cheques still outstanding on 31 March 2026 were to be cancelled and deemed unclaimed, on that report's own terms. Whatever else a percentage measures, it does not measure whether the cheque was opened.

None of this is an argument for despair about insolvency reporting. The court record is unusually good, by the standards of anything else written about a failed crypto venue: a status report is dated, attributed, filed under penalty and written for the people whose money is at stake. What it will not do is hand over a single number, because the process it describes does not produce one until it ends, and closing that gap with an estimate is how the six deleted figures came to exist.

For anyone with a live claim in one of these estates, the practical shape of the work is small and unglamorous. Find the claims agent named by the court rather than a search result. Read the last status report rather than an article about it. Confirm which legal entity and which cohort the claim sits in, because both change the entitlement. And check whether anything has already been sent.

Sources

• Bankruptcy docket, Voyager Digital Holdings, No. 22-10943, Bankruptcy Court for the Southern District of New York — the petition date, the last date to file claims, the initial distribution at Doc 1459 and the reopened in-kind withdrawal period at Doc 1480, the sequence of status reports, and the date of the last known filing, read 23 August 2026.
• Plan Administrator's Eighth Status Report to Creditors, Doc 1785, filed 25 September 2024 — the 70 per cent combined total and its wording, the second distribution at 34.28 per cent, the cheque counts and the undeposited totals, and the refusal to speculate on eventual recoveries.
• Bankruptcy docket, Celsius Network LLC, No. 22-10964, Bankruptcy Court for the Southern District of New York — the distribution notices at Docs 7871, 8188 and 8314, which carry the cumulative-recovery concession quoted above. Doc 4298, the notice of effective date, is not cited: it returned 404 from the free docket mirror and the search surface refused automated access on the date below, so the figure it would have carried is not printed. Read 23 August 2026.
• Voyager Plan Administrator, status reports to creditors — the Tenth Status Report of 27 February 2026. It counts the creditors holding undeposited second-distribution cheques as at 10 February 2026, and sets the 31 March 2026 cancellation date. Read 23 August 2026.
• Celsius Distributions, claims administrator, unclaimed property procedures — the order of 12 February 2025, the two fixed deadlines, the rolling e-mail clocks, and the disposition of forfeited funds. Read 23 August 2026.
• Kroll Restructuring Administration, claims agent to the Terraform Labs wind-down trust — the crypto-loss claims bar date of 16 May 2025 at 11:59 p.m. Eastern, and the quoted consequence from the bar date order. Read 23 August 2026.
• Kroll Restructuring Administration, BlockFi distributions case site — the Convenience Class entitlement and its cap, the currency of payment, the default payment rails and the reversion of unclaimed funds. Read 23 August 2026.
• Federal Reserve Bank of Chicago, Chicago Fed Letter No. 479, May 2023, Radhika Patel and Jonathan Rose — the withdrawal-pause and petition dates, the customer counts drawn from Official Form 206, and the absence of any recovery figure. Cited here for what it contains and for what it does not.

Frequently asked questions

Why is there no average recovery rate for failed crypto platforms?

Because there is no defined population to average over. Each estate uses its own claim pool, its own valuation date and its own definition of what counts in the numerator, and several estates are still distributing, so any average would mix running totals with final outcomes. A cross-case figure also hides the thing that decides an individual outcome: which entity the claim sits against, which cohort it falls into, and whether the money was ever collected. Where a number circulates without a stated denominator it cannot be checked, which is the property that lets it spread.

What does a 70 per cent recovery figure actually measure?

In the Voyager estate it measures a running total against allowed claims, for one legal entity, for the creditors who collected. The Plan Administrator's Eighth Status Report to Creditors states that combining the two distributions, creditors of Voyager Digital, LLC who received the second distribution had received a total of 70 per cent of their allowed claim as at September 2024. Each clause narrows it. Distributions to date is not a final outcome, one entity is not every Voyager company, and creditors who received the second distribution excludes roughly 148,000 uncashed cheques worth about 83 million dollars on the same date.

Does a claim fixed in dollars protect a crypto depositor?

It protects the dollar amount and not the asset. Celsius claims were fixed at petition-date value on 13 July 2022 and Voyager's petition date was 5 July 2022. A claimant paid in full at those values receives the dollars owed and keeps none of the price movement since, so the same outcome reads as complete recovery in one unit and a total loss of exposure in the other. Both statements are true, and estate reporting uses the first.

Can a distribution be missed even when the money has been sent?

Yes, and it happens at scale. The Voyager estate mailed approximately 495,000 cheques for around 587 million dollars in its second distribution, and as at 23 September 2024 approximately 148,000 of them, worth about 83 million dollars, had not been deposited: about 30 per cent of the cheques and roughly 14 per cent of the round by value. Almost 90 per cent of the outstanding cheques were for less than 1,000 dollars. In the Celsius estate, unclaimed distributions are not held indefinitely: under procedures the court approved on 12 February 2025, forfeited funds are redistributed to other creditors who have valid claims or used to fund wind-down expenses.

How can a failed platform still advertise insurance on its website?

Nothing takes a marketing claim down. Reading coinloan.io on 23 August 2026 returns a site advertising a certified custodian with $250 million insurance and a badge asserting a European financial licence, while its own licences page lists exactly two documents: a FinCEN money services business registration that states on its face that FinCEN does not verify what registrants submit, and a European Union trade mark certificate from the intellectual property office in Alicante. The site footer carries a copyright ending in 2023. This is why an advertised protection cannot be assessed by reading the advertiser.

What is the difference between a distribution notice and a payment?

A notice records an intention and a later filing records the event. In the Celsius case, Doc 7871 of 27 November 2024 said the debtors intended to commence a distribution in the near term, and commencement is evidenced only by Doc 8188, which records that the Plan Administrator commenced the second distribution on or around 6 December 2024. Reading a notice as a payment converts a projection into a recovery, which is the most common way a wrong figure acquires a date and starts to look sourced.

Related resources

• Crypto lending risks and insurance — the cluster guide: the four distinctions that decide whether a protection is enforceable before anything goes wrong.
• Exchange security checklist — what to verify on a venue before depositing, including how far a published reserves proof actually goes.
• Custodial and non-custodial lending compared — most estates on this page held customer assets, and the comparison sets out what changes when nobody does.
• Crypto lending platforms — the hub for the lending topic, and the place the surviving venues are described in detail.

Financial Disclaimer

This content is not financial advice. All information provided is for educational purposes only. Cryptocurrency investments carry significant investment risk, and past performance does not guarantee future results. Always do your own research and consult a qualified financial advisor before making investment decisions.