The $2.3 Trillion Ghost: Inside Hyperliquid's Anthropic Pre-IPO Perpetual

At 04:12 UTC on 9 September, a contract with the ticker ANTH printed $2,327.

Nothing about that number is remarkable in isolation. Crypto prints six-figure nonsense every cycle. What matters is what the number implies. ANTH is a perpetual futures contract whose underlying is not a token, not a basket, not an index of liquid assets — it is the equity of Anthropic, a company that has never filed an S-1, has never maintained a public shareholder register, and has never consented to being repriced by an on-chain order book. Multiply $2,327 by the roughly one billion shares the contract's design appears to assume, and you arrive at $2.327 trillion.

That is the headline everyone repeated. It is also, almost certainly, not a valuation.

The snapshot that circulated alongside the print listed three numbers: a mark of $2,327, 24-hour volume of $14.32 million, and open interest of $28.25 million. Three data points. No funding rate. No unique trader count. No oracle specification. No settlement rule. No mention of who deployed the market, under what authority, or against what reference price the contract would eventually resolve.

I have spent nine years watching crypto markets build instruments faster than they build the infrastructure to price them honestly. This is the cleanest example I have seen since 2021. A $2.3 trillion mark, backed by an order book that turns over less than one ten-thousandth of a percent of that notional in a day, is not price discovery. It is a quote wearing the costume of a price — and the difference between those two things is the entire story.

What HIP-3 Actually Ships

To understand how a private company's equity ended up on a decentralized derivatives venue, you have to separate the plumbing from the product.

Hyperliquid operates as a purpose-built L1 with an on-chain central limit order book. That architecture matters for what follows. Unlike an AMM, where price is a deterministic function of reserves, a CLOB concentrates risk at the top of the book. The last trade is not the price. The last trade is the last trade. The price is whatever the next marginal order does to a finite stack of resting liquidity — and when the stack is thin, the last trade becomes a wildly unrepresentative sample of anything.

Standard Hyperliquid perps settle against a validator-submitted price feed: a median of independent quotes, which gives the index an external anchor. Funding then does the work of tethering the perpetual to that anchor. If the perp trades rich to the index, longs pay shorts, and the carry incentivizes arbitrageurs to sell the perp and buy the spot. That mechanism is the whole reason perpetuals have been able to replace dated futures for most of crypto. Remove spot, and you remove the anchor. Remove the anchor, and funding becomes a number that references itself.

HIP-3 is the framework that allows third-party builders to deploy their own perpetual markets on the venue, with custom oracle definitions, a fee split arrangement, and a Dutch-auction process for ticker allocation. The design intent is permissionless financial product creation. The design consequence is that the party deploying the market also defines what the market is priced against. That is a meaningful concentration of power, and it is the first thing I would audit.

The ANTH market appears to have been deployed by a builder operating under the name Entropy. I want to be precise here: I could not independently verify the deployer's identity, the oracle specification, the index composition, the settlement methodology, or the legal wrapper — if any — from public documentation. The source snapshot disclosed none of it. Those are not minor omissions. They are the load-bearing beams.

Pre-IPO exposure is not a new idea on-chain. It is a graveyard. The 2021 cycle produced tokenized equity experiments at several venues; most retreated under regulatory pressure, and the ones that survived narrowed their scope to public-company shares held by a custodian. The pattern is consistent: the technology to represent a private share on a ledger is trivial. The technology to settle that representation is not, because settlement requires a registrar, a custodian, a transfer agent, and a court that will enforce the transfer. None of those are on-chain. That is not a limitation of the code. It is a limitation of the world the code is trying to describe.

Which is why the interesting question about ANTH is not "can it be built." It clearly can. The interesting question is what, precisely, was built — and whether the thing that was built can be resolved without a counterparty that exists in the physical world.

The Arithmetic of a $2.3 Trillion Quote

Let me lay out what the disclosed numbers actually support.

| Metric | Value | |---|---| | Mark price | $2,327 | | Implied fully-diluted valuation (at 1.0B shares) | ~$2.327T | | 24h volume | $14.32M | | Open interest | $28.25M | | OI as % of implied valuation | 0.00121% | | 24h volume as % of implied valuation | 0.00062% | | Implied average holding period (OI ÷ daily volume) | ~1.97 days |

The first comparison that matters is vertical: how does this compare to a mature perpetual market?

Take BTC. Across major venues, aggregate BTC perpetual open interest typically runs in the neighborhood of 2–3% of Bitcoin's circulating market capitalization. ETH is comparable. These are not precise constants — they drift with leverage appetite, basis, and volatility — but the order of magnitude is stable at roughly 10⁻².

ANTH's ratio is 1.21 × 10⁻⁵.

That is roughly three orders of magnitude thinner than a major perpetual, measured the same way. Daily turnover tells the same story with a slightly wider gap: BTC perps frequently turn over 5–10% of market cap per day, while ANTH moves 0.00062% of its implied valuation. Divide one by the other and you get a turnover ratio somewhere between 8,000 and 16,000 times lower.

The instinct at this point is to say "illiquid, therefore unreliable, therefore ignore it." That instinct is correct but shallow. Illiquidity does not make a price wrong. Illiquidity makes a price cheap to move, which is a different and more dangerous property. An incorrect price can be arbitraged away. A cheap-to-move price can be moved, held, and used as a reference by people who never traded it.

Here is where my 2020 work becomes relevant. While finishing my undergraduate thesis, I manually traced $45 million of liquidity flow through Uniswap V2 across roughly 12,000 Ethereum transactions, looking specifically at how slippage tolerance settings interacted with arbitrage efficiency. The finding that got 500 people to follow me was unglamorous: in low-depth pools, the optimum trade size for an arbitrageur is smaller than the size that appears profitable at the quoted price, because the trade itself moves the price. The slippage is not a fee you pay. The slippage is the mechanism by which your own order destroys the opportunity you were trying to capture.

That same physics governs a thin CLOB. If the visible depth within 50 basis points of mid in ANTH totals less than $100,000 in notional — which is entirely plausible for a book whose entire open interest is $28.25 million spread across an unknown number of positions — then a single $250,000 market order, representing just 1.7% of a day's volume, wipes the book and reprices the contract by several percent. Reprice it several times in the same direction and you have manufactured a move that a charting algorithm will read as a trend.

A $2,327 print on $28 million of open interest is not a market's opinion about Anthropic. It is the residual of whatever the last marketable order happened to be.

The Oracle Ouroboros

Now the structural problem, which is the part I think will matter in hindsight.

In a conventional perpetual, the funding mechanism is a closed loop with an outside reference. Perp price drifts above index → funding goes positive → longs pay shorts → arbitrageurs sell perp, buy spot, collect the basis → perp price converges. The outside reference is essential. Without it, funding is just a tax that one side of a two-sided book pays to the other, and the equilibrium price is whatever the participants agree it is.

For a pre-IPO equity perpetual, there is no spot. Anthropic shares do not trade continuously. They change hands in negotiated secondary transactions, at staged primary rounds, and occasionally in tender offers — events that occur on the order of quarters, not blocks. The most frequently cited reference points are round valuations, which by construction are backward-looking, priced at the last primary, and often accompanied by structure (liquidation preferences, participation rights) that make the headline number a poor proxy for common equity value.

So the oracle for ANTH must be one of a small number of things:

1. A third-party private-market data feed — a vendor publishing a periodic mark for Anthropic common or preferred stock.
2. A stale primary-round reference — the last disclosed post-money valuation, updated when news breaks.
3. A venue-internal mark — the order book's own mid, last trade, or a time-weighted average thereof.
4. Some hybrid, weighted between an external reference and the venue's own print.

Option one is the only one that produces genuine external price discovery, and its update cadence would be measured in weeks. Option three is self-referential: the contract is priced off the contract, funding is computed against a price the contract produced, and the entire mechanism becomes a mirror. Option two collapses toward option three between events, because a quarterly anchor cannot discipline a market that clears every block.

A quarterly mark cannot govern a millisecond venue. The anchor exists, but it is asleep most of the time, and in the gap the order book is free to price anything.

This is where I would spend my first hour if I had code access. HIP-3's builder-defined oracle model means the market's deployer specifies the reference. That is a legitimate design for a permissionless system. It is also a single point of control over the number that determines every liquidation, every funding payment, and every margined position's survival. In the 2021 NFT flare investigation, I analyzed 8,500 secondary sales on OpenSea for a prominent PFP collection and found that roughly 40% of reported volume traced back to five interconnected wallets. The manipulation did not require hacking anything. It required only that the venue's displayed price be computed from trades the manipulator controlled. The cheapest attack on any market is not against its liquidity. It is against its definition of price.

I do not have evidence that ANTH's oracle is venue-internal. I also do not have evidence that it is not, because the specification is undisclosed. In an audit, the absence of documentation is itself a finding.

Carry Is the Whole Trade

There is a second-order problem that almost nobody discussing the $2.3 trillion number has priced.

Perpetual funding is usually presented as a small percentage — 0.01% here, 0.03% there. On a five-minute scalp, that is noise. On a two-year thesis, it is the trade.

The conversion is mechanical. With standard eight-hour funding intervals, the annualized rate is the per-interval rate multiplied by three (intervals per day) and by 365:

Annualized funding = per-interval rate × 1,095

A funding rate of 0.05% per eight hours — which reads as trivial on a venue's UI — annualizes to 54.75%. A rate of 0.10% annualizes to 109.5%.

Now extend that over the horizon the instrument actually implies. If Anthropic's IPO is 24 to 36 months out, a long held continuously at 50% annualized funding pays cumulative carry of roughly 171% to 348% of notional on a continuously compounded basis. The carry alone exceeds the terminal price move in most realistic scenarios. A long who is right about direction can still be wiped out by the financing.

This is not a theoretical concern. It is the central design flaw of any multi-year perpetual: perpetuals are engineered for continuous financing of short-duration positions, and they are being used here to express a multi-year view. The mismatch between instrument duration and thesis duration is not a bug in the pricing. It is the pricing.

The Terra collapse is instructive. In May 2022, I rebuilt my research framework around stablecoin reserve auditing and tracked roughly $2 billion of outflows from Anchor Protocol in real time, publishing a predictive alert 48 hours before the main crash. The saving grace of that episode, analytically speaking, was that Anchor's yield was explicit. Everyone could see the 19.5%. The failure mode was not hidden; it was disclosed and ignored, because the number was large enough to be seductive and the mechanism was boring enough to skip. Perpetual funding on a pre-IPO market is the same structure with better camouflage. The number is small per interval and catastrophic per year, and the mechanism is buried one layer down.

If ANTH's funding is running meaningfully positive and persistent, the market is telling you something specific: longs are crowded, and the carry is being paid by people who expect to exit before it compounds. That is a description of momentum flow, not of capital formation.

The Two-Day Thesis on a Two-Year Instrument

Here is the most elegant inconsistency in the disclosed data.

Open interest of $28.25 million against 24-hour volume of $14.32 million gives a ratio of 1.97. The naive reading is that the book turns over every two days, implying an average holding period of roughly 48 hours.

Compare that to what the instrument is. A perpetual on a pre-IPO company has a payoff tied to an event that has not been scheduled. The realistic tenor is measured in quarters. Call it 24 months, or roughly 730 days, to be generous about how quickly Anthropic could plausibly list.

One point nine seven divided by 730 is 0.27%.

The average participant in this market holds a 24-month thesis for roughly 48 hours — 0.27% of the instrument's life.

This is not a market forming a view on Anthropic's terminal value. It is a market where the marginal participant is running a two-day trade and being quoted a two-year liability. Those two facts cannot both be load-bearing. Either the holding period estimate is wrong — possible, if a small number of large, sticky positions are offset by a large number of rapid scalps — or the participants are systematically underestimating the duration of what they own.

I have seen this exact pattern before, in a different asset class. Earlier this year, I designed an experiment in which autonomous AI agents executed 10,000 micro-transactions on a new L2 to measure gas fee volatility. The dataset ran to terabytes, and the finding that stuck with me was not about fees. It was that algorithmic flow in thin markets produces predictable liquidity gaps — moments where the book is reliably empty because every agent has exhausted the same signal and stepped away simultaneously. In the preprint for that whitepaper on algorithmic market microstructure, I modeled these gaps as a function of agent population homogeneity. The more agents run the same strategy, the more synchronized their absence becomes.

A pre-IPO perp with a two-day holding period and no fundamental anchor is a structurally attractive venue for exactly that kind of homogeneity. If the dominant strategy is "buy the narrative, ride the mark, exit before funding gets expensive," then the liquidity gap at the exit is not a risk. It is a certainty with an unknown timestamp.

Follow the Data, Which Means the Wallets

Everything above is inference from three disclosed numbers. Inference is where I start, not where I stop.

When I investigated the 2021 NFT volume anomaly, the method was not statistical. It was forensic. I pulled every secondary sale, clustered wallets by funding source, gas price fingerprinting, and temporal correlation, and looked for graphs that were internally circular — wallets that bought from each other at escalating prices with no third party in the loop. Five wallets. Forty percent of volume. The tell was not the price. The tell was the topology.

The same method applies to ANTH, and the required data is all on-chain. Here is what I would pull, in order:

Wallet-level PnL decomposition. A perpetual market's volume figure is meaningless without knowing how many addresses produced it. If $14.32 million of daily volume comes from 40 addresses, the market is a private negotiation with a public tape. If it comes from 4,000 addresses at $3,600 each, it is retail speculation at scale. The psychology, the liquidation cascade risk, and the oracle attack surface are completely different. This is the single most important number in the snapshot, and it is missing.

Funding rate history. The annualized carry, plotted against price. If funding is persistently positive while price is flat or rising, longs are paying to stay. If funding flips negative while price holds, someone with size is hedging — or someone with size is short and being squeezed. Either way, the funding curve is more informative than the price curve.

One-sidedness of open interest. Hyperliquid's public data allows reconstruction of position concentration. If open interest is 80% long, the liquidation engine has a directional liability. That is not a market-making observation; it is a systemic-risk observation, because of who the backstop is.

Order book depth at 1% and 2% from mid. This is the number that determines whether the $2,327 mark survives a $100,000 order. I would snapshot depth at multiple points in the day, because thin books have diurnal structure — and if depth evaporates at specific hours, that tells you who is asleep.

Unique holder growth versus volume. In my NFT work, the metric that separated real adoption from manufactured hype was unique-holder count, not volume. Volume can be faked by two wallets. Unique holders cannot. If ANTH's holder count is growing while volume declines, that is accumulation. If volume is growing while holder count is flat, that is the same 12 wallets arguing with themselves.

The deployer's wallet behavior. HIP-3 deployers earn a share of fees. If the deployer's address is also a large counterparty in the market, the arrangement deserves scrutiny. I am not alleging anything. I am noting that the incentive structure would make such an arrangement rational, and that the data to confirm or refute it is public.

Transparency is the only security. Everything on that list is verifiable by anyone with an archive node and a weekend. The fact that none of it appeared alongside the $2.3 trillion headline is the actual news here.

The Settlement Cliff

Suppose every inference above is wrong. Suppose the oracle is robust, the depth is honest, the holders are thousands, and the carry is negligible. There is still a structural problem that no amount of market quality solves.

What happens at the IPO?

A perpetual has no expiry. That is its defining feature and, for this use case, its fatal one. Pre-IPO perpetuals on private equity must have a conversion event — a defined moment at which the contract stops being perpetual and starts being a claim on a specific number. That number has to come from somewhere. If it comes from the IPO print, then everything depends on whether the IPO print is a valid measure of the equity the contract references.

Run the arithmetic.

At a mark of $2,327, if Anthropic lists at a $500 billion valuation — which would be a spectacular outcome by any conventional standard, and a multiple of its last reported private rounds — the implied per-share value is roughly $500, against a mark of $2,327. That is a 78.5% drawdown on the settlement print.

At a $183 billion valuation — within the range that has been reported for its most recent private financing discussions — the per-share value is roughly $183. That is a 92% drawdown.

Even in a scenario where Anthropic becomes one of the most valuable companies on earth and lists at $1 trillion, the implied settlement is $1,000 per share against a $2,327 mark. A 57% loss on a thesis that turned out to be correct.

This is the definitional risk of paying any price for any asset: the price embeds a terminal assumption, and the terminal assumption is not yours to control. What makes ANTH unusual is the size of the gap between the mark and any plausible anchor. The 2024 IBIT/GBTC episode taught me how much value lives in this gap. In the first month of spot Bitcoin ETF trading, I quantified a 0.3% arbitrage opportunity created purely by settlement timing differences between two funds holding the same underlying. Three-tenths of a percent — from nothing but a clock. Scale that logic to a market where the settlement event is 24 months away and the reference is undefined, and the clock stops being a fee and becomes the entire risk.

And note the asymmetry. Longs at $2,327 face a settlement cliff measured in tens of percent. Shorts at $2,327 face the carry problem described earlier. There is no clean side. That is what an unanchored market looks like from the inside: not a fat pitch, but a pair of liabilities with different expiration dates.

The Backstop That Marks Itself

One more structural consideration, and this one is specific to the venue's architecture.

Hyperliquid runs a liquidation engine with a vault backstop. In a normal market, this is a sound design: liquidations are absorbed by a capital pool that earns the liquidation premium, and the pool's exposure is to assets with deep spot markets that can be hedged or unwound. The vault is not taking a view. It is providing a service.

Now put the backstop behind a market where open interest is $28.25 million against an implied valuation of $2.327 trillion, where the mark may be venue-internal, and where there is no spot market in which to hedge.

If a large long is liquidated, the engine closes the position. But closing a position requires a counterparty. In a book whose depth within 1% might be a few hundred thousand dollars, the liquidation itself is the market impact. The backstop absorbs the residual. And here is the problem: the backstop now holds a position in an instrument whose price is defined by the venue's own mark.

The vault's solvency becomes a function of a number the venue publishes. If the mark is high, the vault looks healthy and the position looks valuable. If the mark is corrected — by an oracle update, a real IPO print, or a governance decision — the vault's equity revalues instantly and the loss surfaces in one block.

I am not claiming this is the current state. I am saying it is the state the architecture permits, and that the size of the tail is proportional to the size of the gap between the mark and the anchor. At a $2.3 trillion mark, the gap is enormous.

Exit liquidity is someone else's entry. In a market this thin, the person selling to you at $2,327 is not making a valuation call. They are finding a buyer. The question you should ask before taking the other side is not whether Anthropic is worth $2.3 trillion. It is who will be there when you want to leave.

The Case That the Perp Is the Honest Price

Now let me argue the other side, because the obvious conclusion is too easy and I do not trust easy conclusions.

Every criticism I have leveled above assumes the private market is the truthful reference and the perp is the distortion. What if it is the reverse?

Private equity marks are terrible. They are quarterly at best, more often semi-annual, and frequently stale by design — a flat mark protects a fund's reported IRR, and an honest write-down invites redemptions and LP pressure. Within a round, the price is negotiated with structure attached: liquidation preferences, participation rights, ratchets, board seats. The "post-money valuation" printed in a press release is a marketing artifact as much as a financial one. Two investors entering the same round can hold economically different instruments at the same headline price.

Against that, a continuous market — even a thin one — produces something private markets do not: a time series. It produces a price at 04:12 UTC and another at 04:13 UTC, and the sequence of those prices carries information about disagreement, about flow, and about how the marginal buyer's conviction changes as news arrives. When a hyperscaler announces a fresh AI capex commitment, or a competitor releases a model that benchmarks well, or a regulator signals interest, the private market does not move for weeks. The perp moves in a block.

That is genuine information gain. It is the same argument the crypto derivatives market made against traditional commodity futures a decade ago, and it was correct then. Continuous price discovery is a public good, and the fact that a market is small does not make its signal worthless — it makes it noisy. Noise can be averaged out. Silence cannot.

There is also a demand-side reality that the private markets structurally cannot serve. Retail and non-accredited capital has essentially no on-ramp to Anthropic. Secondary platforms exist, but they are gated, illiquid, expensive, and slow. A leveraged perpetual on a decentralized venue is the only instrument on earth that lets a trader with $500 express a view on whether the leading AI safety lab becomes a trillion-dollar company. Whether that is good is a separate question. Whether that demand is real is not in doubt.

And the magnitude should not be dismissed purely on reflex. The $2.327 trillion number is extraordinary, but so is the range of published views on Anthropic's future. Over the last eighteen months, reported private-market valuations have moved across a range spanning more than an order of magnitude, and commentary from serious institutional analysts has spanned even wider. When the fundamental dispersion is that large, an extreme print is not automatically an error. It may be a tail opinion finding expression.

Correlation is not causation, and a mark is not a valuation. But the inverse also holds: a mark that disagrees with consensus is not automatically a manipulation. Sometimes the outlier is the market telling you the reference is wrong.

Here is where I land, and it is not comfortable in either direction.

If the perp is a genuine expression of demand for pre-IPO AI exposure, then the $2.327 trillion number is a signal about scarcity, not about Anthropic. It is telling you that the supply of instruments allowing anyone to hold AI's upside is so constrained that a synthetic claim on a private company, offered on a thinly traded venue with undefined settlement, can be bid to a valuation that no fundamental model supports. That is a statement about market structure, not about the AI industry's prospects.

And it is a statement with a deadline. If the supply of AI exposure normalizes — through an IPO, through a listed competitor, through a wave of tokens claiming to represent AI equity — the scarcity premium compresses. The velocity of that compression is determined by the carry, and the carry is running every eight hours.

The Part Nobody Has to Audit Because It Might Not Exist

One more layer, and then the signals.

Anthropic has not authorized this market. As far as the disclosed record shows, the company has no relationship with it, receives no economics from it, and has no governance rights over it. That is not unusual for crypto, where synthetic exposure to almost anything has been built without permission. But it introduces a category of risk that quantitative analysis cannot price, because it is binary and exogenous.

If Anthropic's counsel sends a cease-and-desist over name and ticker usage, the venue's rational response is to delist, rename, or force-settle the market. If it does not, the market continues. There is no middle path where the contract keeps trading exactly as before. Every position's terminal value depends on a legal question that no funding rate captures.

The same logic applies to securities regulation. A perpetual referencing a private company's equity has characteristics that regulators have historically treated as security-based swap exposure — a leveraged, referenced claim on an issuer's value, offered for trading without registration, to a permissionless global audience that is not restricted from US persons at the protocol level. The venue may take the position that it is a neutral piece of software and that its builders are independent. That argument has been made before, and it has a mixed record.

I do not think this is the most likely near-term outcome. I think it is the tail that matters, because it is the one tail that resolves instantly and without a price path. I learned that lesson the hard way in May 2022: the risk you can model is never the risk that takes the money. The risk that takes the money is the one that changes the rules of the market itself.

Nor is the gaming analogy as far-fetched as it sounds. The reason in-game asset economies keep failing is not technical. It is that the issuing studio can mint arbitrary supply whenever it wants, which destroys the credibility of every existing holder. Apply that to pre-IPO synthetics: a venue can create an unlimited number of contracts referencing a company that never agreed to be referenced. The supply is permissionless. The underlying is not. That mismatch is the same structural flaw, wearing a different costume, and it has the same failure mode — holders discover that what they own is only as good as the issuer's willingness to keep honoring it.

Signals to Watch Over the Next Seven Days

Three disclosed numbers are not enough to build a position on. They are enough to build a watchlist on. Here is what I am monitoring, and what each signal would mean.

Annualized funding rate on ANTH. This is the single most diagnostic number available. If funding runs persistently above roughly 50% annualized while price holds flat, longs are paying heavily to stay, which means the marginal buyer expects a fast exit. That configuration has one characteristic outcome. If funding hovers near zero, the position is being held, not flipped — which changes the interpretation entirely. Cross-reference the funding curve against the price curve, not against the news flow.

Order book depth at 1% and 2% from mid, sampled at fixed hours. I want to know whether the $2,327 mark is defended by six figures of resting liquidity or by four. Depth that evaporates at specific hours tells you the market is one or two participants wearing a crowd costume.

Unique trader count and holder growth. Volume is a headline. Holder count is a census. If the address count is flat while volume climbs, the market is circular. If it is climbing while volume is flat, someone is accumulating quietly — which is the configuration I would pay attention to.

Any statement from Anthropic. Not a rumor, not a tweet from an adjacent employee, an actual legal or corporate communication acknowledging or repudiating the market. That single document is worth more than any technical analysis of the chart, because it resolves the binary risk described above in one direction or the other.

Follow-on HIP-3 deployments. If builders broadly replicate this structure for other private AI, space, or fintech names in the coming weeks, the scarcity premium that may be inflating ANTH compresses by construction, and the dynamic shifts from "the only way to trade Anthropic" to "one of many ways to trade private equity." That transition is where the real repricing happens — and it will move faster than any of the people currently holding long positions at $2,327 expect.

None of these signals tells you what Anthropic is worth. That number does not exist in any honest form yet. What they tell you is whether the market that claims to know it is a market or a mirror.

Code does not care about your feelings. Neither does a funding curve. The only question that has ever mattered in a thin market is who is on the other side — and $2.3 trillion, printed on $28 million of open interest, tells you almost nothing about the answer.