Huobi HTX published its 48th consecutive monthly Proof of Reserves this October. The headline is clean. Every major asset sits above 100%: Bitcoin at 104%, Ethereum at 102%, Dogecoin at 102%, the exchange's own platform token at 103%, TRON at 108%. Full redemption promised. Instant withdrawals guaranteed. A Merkle tree to prove it all. Then you reach the annex, and USDT and USDC have been folded into a single line item labeled 'USDs.' That is not a rounding choice. That is a transparency subtraction dressed as a housekeeping note. In a market where every exchange now publishes a reserve report, the question stopped being 'do they have the assets' and quietly became 'who is allowed to check.' HTX answered the first question and declined the second.
The Proof of Reserves ritual has a short and unhappy lineage. Before November 2022, almost no exchange published one. FTX did not. FTX also did not hold the customer assets it claimed to hold, and the gap between those two facts vaporized roughly eight billion dollars in customer value and detonated a contagion that took down BlockFi, Genesis, and a dozen smaller lenders. In the wreckage, Proof of Reserves became the industry's confession booth. Binance rolled out a zk-SNARK version. OKX published a Merkle tree alongside an independent verification tool. Bitget advertised its over-collateralization. Kraken had been doing it quietly for years. The genre was born not from innovation but from a near-death experience, and it has been normalizing ever since.
Huobi HTX enters this story with an unusually tangled past. The exchange launched in 2013, survived China's 2017 crackdown, and built one of the largest retail books in Asia. In 2022, control changed hands. In 2023, the brand was renamed HTX. Around the same period, Justin Sun, the founder of TRON and a figure the U.S. SEC charged in 2023 with fraud and market manipulation, became the exchange's most visible associated party. The public record on who exactly owns and controls HTX is thin. The exchange now operates across offshore jurisdictions, serves a global retail base, and presents itself through monthly reserve reports. Its self-description is that it is one of the earliest exchanges to disclose reserves continuously, now 48 months running.
A Merkle tree, for the uninitiated, is a hash structure that lets a large dataset be verified piece by piece. Each user's balance is hashed into a leaf; leaves combine into branches; branches combine into a single root. If your balance is in the tree, you can prove it with a short path of hashes. It is elegant, efficient, and, crucially, only as honest as what gets loaded into it. Decoding the signal from the blockchain noise begins with understanding that the tree verifies inclusion, not solvency. Those are different promises, and the gap between them is where depositors get hurt.
Start with the numbers, because the numbers are the part HTX wants you to see. Bitcoin reserves at 104%. Ethereum at 102%. Dogecoin at 102%. The HTX platform token at 103%. TRON at 108%. The aggregate band lands between 102% and 109%. On a press release, that reads as safety. In risk terms, it reads as a fuse.
Here is the arithmetic the marketing never runs. A reserve ratio of 102% means the exchange holds two cents of buffer for every dollar of user liability. If the reserve asset falls 2% while the liability stays fixed, and liabilities are denominated in dollars or dollar-pegged units, the buffer is gone. If the asset falls 10%, the exchange is at roughly 92% coverage, meaning it cannot return every customer's money at once. ETH at 102% and DOGE at 102% are, in a 10% drawdown, insolvent on paper. The crypto market produces 10% drawdowns the way the Pacific Northwest produces rain. Bitcoin routinely moves 10% in a week; altcoins move 10% in a day. A reserve buffer of 2% is not a cushion. It is a speed bump placed at the edge of a cliff.
When I audited twenty failed protocols after the 2022 collapse, one pattern repeated with almost mechanical reliability: the entities that failed were not the ones with the lowest reported ratios. They were the ones whose reported ratios were structurally incapable of capturing the risk that eventually killed them. A ratio is only as good as the liability definition behind it. HTX is reporting a healthy ratio against a liability base that it alone defines and that it never discloses in absolute terms. That is not a measurement. That is a claim.
Now the harder problem, the one the 48-month streak is designed to bury. A Merkle-tree Proof of Reserves proves one thing and only one thing: that for every user balance loaded into the tree, there exists a corresponding entry on the reserve side. That is a statement about the numerator and the denominator matching. It is not a statement about the exchange's total obligations.
Consider what the tree cannot see. It cannot see liabilities that were never loaded: over-the-counter borrowings, off-book credit lines, tokenized promises to institutional counterparties, or the quiet reuse of customer collateral in a lending desk. FTX's balance sheet looked healthy until the moment someone asked where the customer money had gone. The Merkle tree has no mechanism to answer that question, because the answer lives outside the tree. It also cannot see the quality of the reserves. A reserve entry that says ten thousand BTC does not tell you whether those coins are owned outright, pledged against a loan, borrowed for the snapshot, or counted twice across two entities. And it cannot see ownership. A reserve you do not own outright is not a reserve. It is collateral, and collateral can be seized, called, or liquidated by whoever actually holds the senior claim.
Which brings us to the most revealing detail in the entire disclosure: the word 'USDs.' HTX has stopped reporting USDT and USDC separately. Since roughly April of this year, according to the disclosure itself, both dollar-pegged assets are merged into a single bucket. On the surface this is presented as a categorization choice. In practice it is a deliberate reduction in granularity, and the reduction matters enormously.
USDT and USDC are not the same asset, and merging them is where transparency goes to die. They have different issuers, different reserve compositions, different redemption mechanisms, different regulatory exposures, and different depeg histories. USDC depegged to eighty-eight cents during the March 2023 banking crisis. USDT has traded at a discount during stress events. If an exchange holds both and reports them separately, a shortfall in one is visible. If an exchange merges them, a shortfall in USDC can be masked by a surplus in USDT. The aggregate clears 100% while the composition rots underneath. This is the same class of maneuver as reporting 'cash and equivalents' to hide that the cash is a promissory note from a related party. HTX's own disclosure language claims that each asset is recorded and verified independently. If that were literally true, USDT and USDC would appear as separate lines. They do not. The disclosure contradicts itself in adjacent paragraphs, and the contradiction is the story.
Then there is the reserve asset that should make any credit analyst stop cold: the platform's own token, listed at 103% coverage. HTX is holding its own IOU as backing for its customers' deposits, and that is a mirror facing another mirror. The value of the HTX token depends on the health of the HTX exchange. The obligations of the HTX exchange are supposedly backed by the HTX token. If the exchange gets into trouble, the token is the first thing to collapse, and it is simultaneously part of the buffer meant to protect depositors. This is not a reserve. It is a circularity.
FTT already ran this experiment for us. The FTX token was backing positions across Alameda and FTX until both discovered that the backing was the liability. When confidence broke, FTT went from the twenties to under two dollars in days, and it dragged the entire exchange down with it. The token did not cushion the fall. The token was the fall. A reserve asset must be liquid, exogenous, and uncorrelated with the issuer's survival. The HTX token is illiquid relative to the book it is meant to cover, endogenous to the issuer, and perfectly correlated with the exact failure it is supposed to insure against. Counting it at 103% is not conservative. It is optimistic fiction with a decimal point.
The reserve line that reads TRON at 108% deserves its own paragraph, because it is the clearest fingerprint of who actually runs this book. TRON is Justin Sun's chain. TRON holds 108% coverage, the highest of any asset in the disclosure, higher than Bitcoin, higher than Ethereum. An exchange does not casually overweight one ecosystem's token above the two largest assets in the industry unless that ecosystem is family. The reserve composition is not a neutral market position. It is a portrait of the ownership structure, painted in asset weights.
This matters for two reasons. First, concentration risk: an exchange whose solvency is heavily tied to a single founder-affiliated token is exposed to that token's idiosyncratic shocks, not just market beta. Second, regulatory risk: Sun's 2023 SEC charges, fraud and market manipulation among them, cast a shadow that does not stop at the TRON blockchain. It reaches any venue where his influence is material. If U.S. or E.U. regulators decide that HTX is functionally a Sun-affiliated entity, the consequences run from market-access restrictions to banking-channel closures, none of which a Merkle tree can hedge.
Now read the disclosure for what is missing. There is no named auditor. The document asserts that reserve information is accurate, clear, and traceable. That is a self-assessment. In accounting, a self-assessment of accuracy is called an assertion, not an audit. The entire value of a reserve report comes from the independence of the party checking it. A report signed only by the party being checked is a mirror, not a witness.
The industry knows this. Mazars and Armanino, two firms that briefly audited exchange reserves after FTX, both pulled back; Armanino cited the strain, and Mazars paused its crypto work entirely. The retreat left a vacuum that exchanges filled with self-published trees. Binance went further and built a zk-SNARK system that lets users verify inclusion without trusting the exchange's math. OKX shipped a verification tool. HTX, by contrast, describes a Merkle tree with no zero-knowledge upgrade, no third-party attestation, and no liability total disclosed. That last omission is the loudest silence in the document. A Merkle tree without a disclosed liability total is a fraction with a hidden denominator. It can prove that some users' balances are covered. It cannot prove that all users' balances are covered, because it never states what 'all' adds up to.
Which brings us to the '48 months' figure, the number HTX leads with and the number that deserves the most skepticism. Continuity is not credibility. A person who lies consistently for 48 months is not more trustworthy than a person who lies once; they are more practiced. The frequency of a disclosure tells you nothing about its quality. HTX has published 48 monthly snapshots, and in every one of those 48 months the same methodological holes were present: no independent auditor, snapshot-only data, aggregation of stablecoins, and a self-referential token in the reserve base. Doing the same incomplete thing 48 times does not compound into completeness. It compounds into habit.
There is a tell in the emphasis itself. Why lead with duration? Because duration is the one metric that cannot be checked against reality. A reserve ratio can be audited. A method can be inspected. '48 months' is a feeling, a warm signal of reliability that costs nothing to assert and resists falsification. When a disclosure foregrounds the metric that is hardest to verify and buries the metrics that are easiest to verify, you are reading marketing, not accounting.
Let me make the window-dressing mechanism concrete, because it is the single most important thing a depositor can understand about monthly reserve reports. Suppose an exchange wants its report to look better than reality. It has a month. Two weeks before the snapshot date, it borrows a large block of Bitcoin from a friendly fund, a market maker, or a related entity, under a short-term arrangement. On snapshot day, the borrowed Bitcoin sits in the reserve wallet. The Merkle tree captures it. The ratio prints at 104%. Two days later, the Bitcoin goes home, and the reserve wallet is thinner than it was before, but the report is already published and the public is reassured. Nothing about this requires fraud in the criminal sense. It requires only a loan and a calendar. The snapshot is a photograph, and a photograph can be staged. Real-time attestation is the antidote; snapshot reporting is the invitation.
There is a deeper version of the same problem: double-counting across affiliated entities. An asset counted once is a reserve. The same asset counted on two balance sheets is a mirage. When an exchange and its affiliated entities, a market maker, a lending desk, a foundation, a chain foundation, share assets, the same coins can appear as reserves on both books. FTX and Alameda did exactly this, and the collapse exposed a balance sheet where assets had been counted in places they never existed. A Merkle tree on one entity's books cannot detect duplication that lives across entity boundaries. It proves the local ledger, not the consolidated truth.
And context sharpens the concern. We are in a bull market. Prices are up, sentiment is euphoric, and the pressure on exchanges to look bulletproof is at its peak. Bull markets are precisely when thin buffers get tested least and hidden most. Everyone feels rich; nobody withdraws; the reserve ratio never gets stress-tested by a real bank run. The 102% buffer looks fine in a rising market because it is never asked to do its job. Thin buffers are pro-cyclical: they feel adequate when prices rise and fail catastrophically when prices fall, at the exact moment liabilities are being pulled. A 102% buffer is a fair-weather friend. It will be there in the sunshine and gone in the storm, which is the only time you needed it.
There is also the question nobody asks because the answer is uncomfortable: who actually verifies these trees? The Merkle structure is designed so that each user can confirm their own inclusion. In theory, every depositor checks. In practice, almost none do. The verification requires downloading a tool, running a hash comparison, and interpreting a result, a workflow that the median retail user will never complete. The result is a transparency mechanism that functions as theater: a proof that is available but never performed. A witness who never takes the stand is not a witness. It is a prop.
Structuring chaos into profitable narratives is my job, and the narrative here is clean and useful to the exchange: 'we disclose every month, we are transparent, we are safe.' The counter-narrative is that disclosure frequency has become a substitute for disclosure quality, and the market has been trained to mistake the two. This is not unique to HTX. It is an industry-wide drift, and HTX is simply the clearest current example.
Here is the contrarian read, the one the bull market does not want to hear. The prevailing narrative treats Proof of Reserves as a net positive for the industry, a transparency upgrade, a post-FTX correction. I think that framing is backwards in one specific and dangerous way. A weak Proof of Reserves is not merely useless; it is actively harmful, because it manufactures confidence without earning it.
A self-published, unaudited, snapshot Merkle tree does not reduce the information asymmetry between an exchange and its depositors. It launders the asymmetry into a reassuring artifact. It gives retail a number to point at, 'look, 104%, they're solvent,' while the actual risk, the hidden liabilities and the pledged reserves, remains invisible. The report does not close the gap. It papers over it with a figure that feels like proof and functions like placebo.
History doesn't repeat, but it rhymes, and the rhyme here is familiar. Every failed exchange of the last cycle published something that looked like reassurance. FTX had a valuation. Celsius had a yield. Luna had an algorithm. The pattern is never the absence of a signal. It is the presence of a signal designed to be reassuring rather than true. '48 months' is that signal in this cycle. The illusion of value in digital scarcity is not confined to profile-picture NFTs; it applies equally to a reserve ratio that proves the numerator while hiding the denominator. When I forecast a 70% correction in low-utility collections back in 2021, the mechanism was the same: a number that everyone pointed to as evidence of value, and nobody checked as evidence of anything else. The honest posture for a depositor is not 'the reserve ratio is above 100%, I am safe.' It is 'the reserve ratio is above 100%, and I still cannot see the liabilities, the audit, or the ownership. The ratio answers a question I did not ask.'
So what comes next? The next cycle of reserve reporting will split along a single fault line: audited, real-time, exogenous-reserve verification on one side, and self-published monthly snapshots on the other. Exchanges that cross that line, with zk proofs, named auditors, disclosed liability totals, and no platform tokens in the reserve base, will earn a trust premium that compounds. Exchanges that keep publishing warm feelings and hidden denominators will keep doing so until the market stops believing the number and starts asking who signed it. The real question for anyone holding assets on a centralized venue is not whether the ratio clears 100%. It is whether the party reporting it is the same party that could be lying. In every meaningful sense, the answer for this disclosure is yes. Read the annex. Count the silences. Then decide who you trust with your keys.