Whose Money Is the Hedge?

A rules-of-the-game piece. Every figure in it comes from the sealed daily record of 14 to 17 September 2026, the first four closed days of the dollar hedge.

There is a sentence in the Core documentation that reads, on first pass, like an error:

The investor funds no part of the hedge, and bears its economics in full.

Read quickly, that sounds like a contradiction, or worse, like the kind of phrasing that exists to make a cost disappear. It is neither. It is two separate facts about two separate questions that happen to sit in one sentence, and the confusion comes from the word "hedge" being used for both of them.

One question is: whose money stands at the exchange, at risk of being frozen, seized or lost if the venue fails?

The other is: whose number moves when the position makes or loses money?

In this fund those questions have different answers. The first is the house. The second is the investor. Everything below is the mechanism that makes that possible, in the order money actually moves through it.

The sentence also needs one qualification, which arrives in section 2 and gets its history in section 10. It concerns $285,187 of inherited investor money that sits in the collateral today. It is named early rather than buried, because a disclosure a reader finds on their own reads as something that was being hidden.

1. What the overlay is

Core holds coins. On 18 September it held 33,091 SOL, 14.83 BTC across two wrapped forms, and 156 ETH, plus dollars. Those coins are inventory for a market-making book, not a directional bet. They are there so the engine can quote both sides of a pair, and the engine's return comes from the spread it earns, not from the direction of the asset.

The problem is that the inventory has a price. A 12% day in SOL moves the fund's NAV by hundreds of thousands of dollars, and none of that movement has anything to do with whether the engine did its job well. On such a day a good week and a bad week look identical, because both are buried under the price of an asset the fund never intended to have a view on.

So the fund holds the mirror image of its own inventory: for every coin it owns, it is short one coin on a perpetual futures contract.

The important detail is the unit. The short is sized in coins, not in dollars. The target is: short exactly the number of coins the fund holds. That is why the mirror does not drift as the price moves. If the fund holds 33,091 SOL and is short 33,091 SOL, a move from $101 to $114 changes both sides by the same amount and cancels, with no action required from anyone.

Two wrapped forms of Bitcoin, held in different pools, are mirrored by one Bitcoin short in proportion to how many of each the fund holds, which on 18 September was 69% of one form and 31% of the other.

On top of the mirror sits a small ladder of far out-of-the-money put options, six of them, on Bitcoin and Ether, struck 10%, 20% and 30% below the market and expiring at the end of October. They are not the hedge; they are insurance on the hedge, for the kind of day when a mirror built on one venue's marks is not enough. They cost $10,225 to buy on 13 September.

Their budget is 1% of NAV a year, enforced on a trailing 365-day basis with a 10% tolerance. It is measured against NAV as it stands when the ladder is sized, not against the opening NAV this article uses for the four sealed days, so today it is $82,264 of premium a year on a NAV of $8,226,417, against a ceiling of $90,491. The ladder currently in place annualises to $79,608, or 97% of the budget. That is close to the line by design rather than by accident, and it means a decision to buy more protection is a decision to raise the budget in public, not a quiet drift upward.

2. The two questions, side by side

A short position at an exchange requires collateral. Not as a fee, not as a payment to anyone, but as a bond: money that sits at the venue so that the venue is willing to carry the position at all. At 2 times leverage, a $5.38M short book requires roughly $2.69M of it.

That collateral is $2,803,985, and it came from three places:

• $1,340,933 of founder capital, transferred into the hedge pool on 13 and 14 September;
• $580,041 of crystallised performance fees, money the management company had already earned;
• $883,010 carried over from the previous hedge structure when that structure was closed on 10 September.

The first two are unambiguously the house's. The third needs a sentence that most funds would leave out, so here it is in the second paragraph of the section rather than in a footnote at the end. Of that $883,010, $597,824 was the house's own equity tranche in the closed pool and $285,187 was senior participations held by investor accounts. That $285,187 was not paid back out to those accounts. It sits in the collateral today, alongside house capital, and it earns its holders nothing while it is there. Section 10 is the history of how it got there.

So the honest version of the claim in the title is narrower than the slogan, and it is this: no investor money funds the running of the hedge, and $285,187 of inherited capital remains posted inside its collateral. Since the structure changed on 13 September, every dollar added to the collateral has come from the house: six top-ups in the four days under discussion, including a fresh $250,000 transfer of founder capital on 14 September. No account has been debited to keep the collateral there, and nobody is charged interest or a financing spread for its use. The inherited $285,187 stays posted on terms agreed with the accounts that held those participations when the old pool closed. It is named here anyway, because an arrangement being agreed is not a reason for a reader to have to discover it.

And the second column of that diagram is just as absolute. The house takes no share of what the hedge earns. There is no first-loss tranche in this structure, no equity slice that keeps the upside of the overlay, no spread skimmed from the result. Every dollar the mirror makes or loses is allocated out to the accounts that hold the coins being mirrored.

The collateral is a bond posted with an exchange. The result is a number in an account. A loss never cuts the first one. It reaches the second one the same night.

3. What happens every minute

A machine re-marks the book once a minute, all day, every day. On each pass it does four things: it reads the oracle price of each asset, advances the mark on the short and writes the change into the hedge journal, checks whether the position has drifted outside its band, and if it has, works a rebalancing order in child clips of up to $50,000, resting on the touch of the real order book before crossing the spread.

In the four sealed days it wrote 29,717 journal rows and traded 23 rebalancing orders in 61 child fills, $2,137,465 of turnover, for $918 of exchange fees and $651 of execution slippage.

None of this is a cash event yet. Through the day the profit or loss of the short is unrealised: it exists in the journal and in the venue's equity, and nothing has been paid by anyone.

4. What happens at midnight

This is the part that makes the whole arrangement work, and it is worth being slow about.

First, what does not happen. The short positions are not closed. Nothing is sold, nothing is bought back, and the mirror does not go flat overnight. The book has stood since 13 September and carries straight through midnight into the next day. What closes at midnight is the accounting day, not the position, and the distinction matters because it is the difference between a hedge and a series of day trades.

This is how a futures exchange has worked for a century: a position can live for years while the money it owes or earns changes hands every single day. That daily settlement is the reason an investor's NAV is honest at the end of each day rather than carrying an unrealised number that might never be paid.

At 00:00 UTC the day is netted. The venue wallet receives the day's variation, which is the sum of everything the position actually earned or owed: the marks, the fees, the slippage, the difference between the venue's mark and the oracle's. Then the net of the day is swept out of the wallet and posted, one entry per account, into each account's own result line. A loss leaves the accounts and makes the wallet whole. A gain leaves the wallet and lands in the accounts.

The wallet statement for those four days reads exactly like that, and it is the cleanest evidence in this piece:

The middle column is why the first and third do not match. Variation is the part of the day that is still a mark when midnight arrives: the marks themselves, the execution slippage, the venue basis. Funding and exchange fees are cash the moment they happen and are already in the wallet, so the amount swept to or from the accounts is the variation plus them. On 17 September that was $716 of funding received less $211 of fees, or $505, and $117,622 less $505 is the $117,116 that reached the accounts.

Two things fall out of that table.

First, a loss never cuts the collateral. It is not a buffer that absorbs losses and shrinks; it is a bond that stands and is made whole every midnight by the people whose coins the position is mirroring. The collateral does grow when the position grows: those four days took it from $2,479,597 to $2,803,985 through six top-ups, all of them house money. What it never does is fall because the hedge had a bad day.

Second, it is the accounts that pay. On 18 September, the day this piece was written, the short was $517,258 under water intraday on a 12% move in SOL. The venue equity fell below the initial margin requirement for a few hours, the effective leverage rose to 2.35 times against a 2.0 cap, and the distance to liquidation stayed at 41%. Those are three different thresholds and it is worth not confusing them: initial margin is what the venue wants before it lets you add to a position, maintenance margin is what it wants before it closes you out, and on that afternoon maintenance was $45,798 against $2.29M of equity. Being below initial margin means no new risk may be added. It is nowhere near being liquidated. At midnight the accounts settled it and the collateral went back to $2,803,985. None of that intraday gap was funded by anyone, because a mark is not a payment.

The other half of that sentence is that the accounts also receive. 15 September was a down day: the mirror made $238,752 and every dollar of it went out to the accounts, with no share retained by the house.

5. Your share, atom by atom

The overlay is one book for the whole fund, but the accounting is per account, and the rule is one line long:

Not its share of NAV. Not its size. Its share of that specific asset's coin count. An account holding 4% of the fund's SOL gets 4% of every SOL line: the mark, the funding, the fee, the slippage. Three separate accounts of the same NAV can therefore receive three quite different hedge numbers, and there is nothing discretionary about it.

This is also why the hedge line on a statement should never be read as a charge. It is an allocation of a result. The entry is posted to the account's result book and never touches its base capital, which means it changes NAV without changing what was paid in. For an account that deposited $100,000, the deposit is still $100,000 after the hedge has moved through it a hundred times.

6. The bill, line by line

Here is one sealed day, 17 September, completely decomposed.

The market repriced the fund's coins by +$110,555. The mirror marked -$110,050 against it. Those two numbers are that close because they are built from the same oracle price and the same coin count; the small gap between them is the only part of the mirror that is not mechanical, and section 8 is about it.

Everything after those two bars is the actual cost of running the overlay:

• the option tail, -$1,764. The puts lose value when the market rises. That is what insurance does on a good day.
• funding, +$716. A perpetual short is paid by longs when the market is positioned long, which it was all four days. Funding was positive for the fund every single day.
• exchange fees, -$211, and execution slippage, -$115. Both come from rebalancing, and both are small because the book barely needed to trade.
• venue basis, -$5,693. The perpetual contract is marked at the exchange's mark price; the fund's coins are marked at the oracle. When those two disagree, the difference is booked as its own line rather than hidden inside the mark. It is the largest and noisiest item on the list, and until this week it was also the most misleading, because one line was carrying two unrelated things.

Writing this article is what made that obvious, so the line was split before publishing it. Execution basis is the gap between the price a rebalancing trade actually arrived at and the oracle at that instant. It is a real, permanent cost, and on 17 September it was +$401, in the fund's favour. Mark basis is the running value of the difference between the perpetual's mark and the oracle. It is not a cost at all: it is a position in a spread, and it reverses when the spread does. On 17 September it was -$6,095.

Across the four days the split is -$141 of execution basis against -$5,940 of mark basis. Almost everything that looked like the overlay's biggest bill is a spread mark that has already partly reversed. From tonight both halves are computed at the settle, stored on the day's record and named on the receipt, so nobody has to take that paragraph on trust.

What reached NAV that day was -$6,561, and every component of it is named above.

The same decomposition for all four days:

One column is missing from that table, deliberately, and it is the one that explains the difference between this measurement and the fund's own books. The "market on the coins" column above prices the holdings the fund had at the start of the day. The engine also trades during the day, and the positions it changes reprice too, which added a further +$12,319 over the four days. The mirror follows the book minute by minute, so it covers that part as well.

Adding it back gives the honest total: +$88,094 of market movement, -$87,022 of hedge, and +$1,072 left over.

7. So what did the exercise achieve?

Over the four sealed days, $558,993 of gross market movement passed through the fund's coins. There are three honest ways to say what happened to it, and a fund quoting only the flattering one should not be trusted, so here are all three.

Day by day, the share of the market's move that did not reach NAV was 99.2%, 98.6%, 96.3% and 94.6%. The median day removed 97.5% of it.

The worst day was 17 September, which let 5.4% through, or $6,046 against the investor.

Summed, the four days left $1,072 in the fund's favour, which is 1.3 basis points of the $7,972,859 the fund opened 14 September with. That last number is the smallest of the three and the least meaningful, because up days and down days cancel inside it: the four residues were +$1,072, +$3,303, +$2,743 and -$6,046, and in absolute terms they add to $13,164, or 2.4% of the movement that went in. Comparing a net against a gross is how a 97.6% becomes a 99.8%, and it is worth naming the trick rather than using it.

For comparison, over the same four days and on the same NAV base, the engine's own work, the spread it earned net of its costs, was +$14,012, or 17.6 basis points. That ratio is the entire point of the overlay. Without it, $14,012 of strategy result would be a rounding error inside $559,000 of price noise, and no investor, including us, could tell a good week from a lucky one.

(One coincidence, before someone finds it and assumes a copy-paste error: the residue on 14 September and the total for all four days both round to $1,072. They are different numbers, $1,072.39 and $1,071.82.)

Protection is not the goal. Legibility is. A market-neutral book exists so that the only thing left moving in your statement is the work.

8. Where it goes wrong

A piece like this is worthless if it only describes the machine working.

Tracking error is real, and it is the largest cost. The gap between the market line and the mirror was -$1,504, +$3,573, -$5,785 and +$506 on the four days. It comes from the band: to avoid trading on noise, the machine only rebalances when the position drifts more than a set amount away from the exact coin count. Inside the band, the fund is slightly over-hedged or slightly under-hedged, and on a day with a large move that costs money.

Until 18 September the band was a flat 5% of exposure with per-asset floors. On 16 September that let the mirror as a whole sit 8.8% above the exact coin count, and the fund paid $5,785 for the privilege of not having traded. Since 18 September the band halves to 2% with floors cut in half whenever an asset has moved more than 3% on the day. The first observation after the change was that same afternoon, a 12% move in SOL, during which the drift stayed at $6,556 against an exposure of $3.77M, or 0.17%. One afternoon is one observation and proves nothing: the rule was chosen from the arithmetic of the four days above, not from a backtest, and the honest way to judge it is the next month of sealed days, in public, including the days it costs more in fees than it saves.

The venue basis is noisy and can be larger than everything else combined. It was -$5,693 on one day and +$2,926 on another. Now that the line is split, the honest reading is narrower: the execution half is small and real, and the mark half is a spread position that has to be watched over months rather than days. If it does not revert, that is a finding about our own mark sources and it will be published as one.

The residue is not evenly distributed, and this is the number we would watch if we were you. Fund-wide the four days netted to +$1,072, but per account the same four days ran from +48 basis points of NAV to -151. Two things cause that, and only one of them is the band.

The first is the band, as above: the accounts the engine trades in most have inventory that changes during the day, and the mirror follows it only when the drift leaves the band.

The second is a mismatch of bases that is worth stating precisely. The overlay's result is attributed to accounts by their share of each coin measured at the day's open, while the book itself is marked minute by minute against whatever the fund holds at that minute. On a day when coins move between accounts, those two are not the same denominator, and the difference lands as dispersion. It is not a rounding artefact and it does not average to zero for an account whose inventory keeps changing. The smallest account in the book, at $8,700, ended the four days $132 behind, which is a large percentage of a small number and a small amount of money. We do not consider either cause acceptable in the long run. The band is being fixed first because it is the larger of the two; intraday attribution is the harder problem and it is on the list.

The tail costs money most of the time. That is the design. It cost $2,236 over the four days and would have to pay for itself in a crash to have been worth carrying. It has not been tested yet in this structure.

Three rules changed tonight, because of this article. Writing down what the machine does exposed three things it did not do. A single broken venue mark used to abort the whole rebalancing pass, so two healthy assets kept their drift for no reason; each asset now skips on its own and says so. A dislocated spread was only stopped at 50 basis points, where the venue is declared broken; a softer gate now holds non-reducing trades at 25, because adding to the short into a dislocation books that dislocation as a real entry. And the collateral was only topped up when liquidation came within 30%, which says nothing about a position that has simply outgrown its bond; it now tops up when notional against posted collateral passes 2.2 times, measured on what is posted rather than on equity, since equity falls intraday and the midnight settle repairs it.

And one thing that is not fixed. The pool the collateral is topped up from is empty tonight. Structural leverage is 1.93 times against a 2.0 design, which is inside tolerance, but it drifted there because each rebalance that grew the book could not draw its share of collateral. The machine now raises this rather than absorbing it quietly, and the answer is a house transfer, not an investor one.

Nothing here has been tested in a real liquidation. The margin guards top up at 30% distance, go reduce-only at 20% and hard stop at 12%, and the closest the book has come is 41%. Guards that have never fired are a design, not a track record.

The arrangement depends on the house being able and willing to keep posting. This is the structural risk of "the collateral is ours, the result is yours", and it belongs in this section rather than in the marketing. The collateral was topped up six times in four days, once with $250,000 of fresh money. A larger book, or a sustained rally against a short, needs more of it. If the house could not or would not post, the overlay would have to be cut back, and the investor would be left holding an unhedged coin position, which is the exact exposure they are here to avoid. Nothing about today's numbers suggests that is close: $2.8M is posted against $5.38M of notional, and the fund's own capital sits behind it. But an investor should know that what protects them is a standing commitment by the manager and not a right they hold.

The wrapped forms of Bitcoin are not hedged as wrapped forms. The fund holds two wrapped Bitcoins and mirrors both with one Bitcoin short. That covers the price of Bitcoin. It does not cover the wrapper: a depeg, a custodian failure or a redemption freeze in either form is a risk the overlay does not touch and cannot touch, because the perpetual tracks the asset and not the token. The same is true of the stablecoin the collateral is denominated in.

9. What the investor bears, and what the investor does not

Plainly, as two lists.

You bear, in full: the mark of the mirror against your own coins; your share of funding, positive or negative; your share of exchange fees and execution slippage; your share of the venue basis; your share of the option premium and its value; and your share of the tracking error, in both directions.

You do not bear: the funding of the collateral standing at the venue, with the inherited $285,187 of section 2 as the stated exception; any financing cost, interest or fee on that collateral; the risk that the venue itself fails while holding it; the cost of building and running the machinery; and the intraday gap between a mark and a settlement, which is carried by the collateral and not by anyone's account.

You do bear one thing that is not on either list, and it is not a cost. The arrangement runs on the manager continuing to post collateral, as section 8 says. That is a commitment, not a contract you hold, and if it ever stopped, what you would be left holding is the coin exposure the overlay exists to remove.

There is one more asymmetry worth naming, because it runs the other way. Since 18 September the idle dollars in Core, which is most of what the hedged inventory sits next to, earn a carry, posted daily, on the three quarters of each account's dollar balance that are not working capital. The rate is currently about 4% a year. The other quarter stays unlent on the venues as the quote side of the market-making book.

That carry is not free money and it is not ours to promise. It is a lending rate: the lent part sits in flexible dollar lending, on an exchange earn product and an on-chain lending market, both redeemable within minutes, which is why it can still back the book. That means it carries the credit risk of those two counterparties and the rate itself moves with the market. Investors keep the yield and also hold that risk. It is disclosed here for the same reason as everything else in this article: a number that appears on a statement without a named source is a number nobody can check.

10. What changed, and when

This has not always been the arrangement, and the previous one was different in exactly the way that matters here.

Until 10 September the hedge was a funded pool. Investors held senior participation units in it: they paid in, they took distributions out, and a first-loss equity tranche absorbed the tail. In that structure, the investor did fund the hedge, and the pool's own equity, not the investor's statement, was where the daily result landed.

That pool was closed on 10 September. Its terminal equity, $883,010, was carried out through the accounts and into the collateral that now stands behind the perpetual book, and the new structure started clean on 13 September: no units, no tranches, no pool the investor is a member of. A short position, a bond posted by the house, and a daily allocation of the result by coin count.

Section 2 gave the split of that $883,010: $597,824 of house equity tranche and $285,187 of senior participations belonging to investor accounts. Here is the sequence, because the sequence is what makes it checkable. On 10 September the closed pool's value was recognised into each account that held a participation, which is why every one of those accounts shows a hedge credit that day. On 11 September the same amount was moved out of those accounts into the pool that became the new collateral, which is why the next day shows an identical debit. Net effect on anyone's NAV across the two days: zero. Net effect on who holds the money: the accounts held a participation before and hold nothing now, and the collateral is larger by $285,187.

That carry-over was agreed with the holders at the cutover rather than imposed on them, and the money stays posted on that basis. It is written down here for the same reason as everything else in this piece: an arrangement a reader has to discover for themselves, after nine sections of arithmetic, reads like something the arithmetic was preparing them for. If those terms ever change, the change will appear in this article, with the date.

The reason for the change is the same reason this article exists. A funded pool with participations requires an investor to understand three things, the pool's own NAV, their units in it, and the relationship of both to their account. A mirror with a daily result requires them to understand one line on a statement.

11. How to read it on your own statement

Your daily Core result has four lines, and now all four should be readable without translation:

• Strategy. What the engine earned, net of its own costs. This is the only line that is supposed to be positive on average.
• Market after hedge. What the price of the assets did to you after the mirror cancelled it. Small by construction. When it is not small, something in section 8 is happening, and it should be explained rather than averaged away.
• Hedge. Your share of the overlay's result for the day, the number this whole piece has been about. Positive on down days, negative on up days, and itemised in the app down to funding and fees.
• Carry. What your idle dollars earned.

If those four lines and your flows do not add up exactly to the change in your NAV, that is a bug, and we would rather hear about it from you than not hear about it.

12. What "sealed" means, since the word is doing a lot of work here

This article leans on the word "sealed" a dozen times, so it should say what the word buys you.

At 00:07 UTC the previous day is closed for every account and written once. The row carries the opening and closing NAV, the strategy result, the costs, the market line, the intraday reprice, the flows, the hedge line, the carry, the residual, the prices used at both ends of the day and the holdings the day opened with. Those fields are hashed together, and the hash is stored in the row.

A sealed row is then never edited. If something is later found to be wrong, the correction does not overwrite it: a separate append-only record is written holding the old hash, the new hash and both versions of the row in full. There have been 1,552 such corrections since 11 June, the most recent on 6 September. That number is deliberately in this paragraph. A record that has never been corrected is either perfect or not being checked.

The four days used throughout this article seal to these digests, each one the hash of that day's 21 account rows, taken in account order:

If any figure in this article is ever restated, the digest for that day changes and the errata record says why. An investor can ask for the hash of their own row for any date and check it against these. Putting that hash on the statement itself, so nobody has to ask, is on the list.

One more piece of borrowed discipline. The Print, our print title, labels every figure by where it came from: OBSERVED for things measured on our own connection, CALCULATED for things derived from disclosed inputs, REPORTED for things established from somebody else's primary documents. On that scale, every dollar figure in this article is OBSERVED, out of the sealed record above. The percentages, the annualised budget and the drift figures are CALCULATED from those numbers, and the arithmetic for each is given in the sentence that uses it. Nothing here is REPORTED, because nothing here comes from anyone but us, which is itself a limitation worth stating: this is a fund marking its own homework in public, and the only defence against that is showing the working.

A bond at an exchange and a number in an account are two different kinds of money. This entire article is the difference between them.

Definitions

• Perpetual future. An exchange-traded contract that tracks an asset's price with no expiry, where longs and shorts exchange a periodic funding payment instead of the contract rolling.
• Mirror, or coverage. Being short exactly as many coins as you hold. Expressed as a coverage ratio, where 1.0 means fully neutral.
• Collateral, or margin. Money posted at a venue so it will carry a position. Not a cost, not a payment, and not consumed by losses as long as those losses are settled in cash.
• Variation. The day's change in the value of a position, paid or received in cash at settlement.
• Funding. The periodic payment between longs and shorts on a perpetual. A short receives it when the market is positioned long.
• Basis. The difference between the exchange's mark on a contract and the reference price of the underlying asset. Real cash, and booked as its own line here rather than folded into the mark.
• Band. The amount of drift from an exact mirror that the machine tolerates before it trades. The only discretionary parameter in the overlay.
• Tracking error. What the band costs, in either direction. The gap between the market line and the mirror.
• Result book. The part of an account that holds outcomes rather than capital. The hedge posts here, which is why it changes NAV and never changes what was paid in.

Related reading: Why Hedge Can Lose Money on the mechanics of the option layer, What Is Capital Protection on what protection does and does not promise, Why NAV Was Up While SOL Was Down for the same arithmetic seen from a single day, and Understanding Units and NAV on how a result reaches your unit price.

Nothing in this article is investment advice. Figures are from the fund's sealed daily record for 14 to 17 September 2026 and are stated as at the date of writing.