"June 2026: The Plumbing Healed Before the Price Did"

The first of a monthly series - and a deliberate choice about what it is, written out so you can hold us to the format.

We are not going to tell you what the market did. BlackRock, JPMorgan, and Wintermute will each write that recap, and write it better - they have larger desks, longer lenses, and more analysts than we ever will. A monthly price summary is a game we would lose, so we are not going to play it. This section plays a different game, the only one where we hold an edge that cannot be replicated from the outside:

Not "here's what the market did." Here's what the market looked like from inside the machine.

We run an engine that quotes, fills, hedges, and rebalances through this market thousands of times a day. That engine emits a continuous stream of telemetry - fill quality, adverse selection, rejection rates, spread capture, latency, capacity - that does not exist on any Bloomberg terminal, because it is a property of participating, not of observing. A terminal can tell you the price of SOL. It cannot tell you whether our resting orders were getting picked off, because they aren't our orders. That gap is the entire reason this report exists.

June is an unusually good first subject, because the price and the telemetry spent the month telling two opposite stories - and the telemetry was, in our judgement, the more honest of the two. This report walks the market down through seven layers, from the one everyone can see to the ones only we can, and ends where every one of these reports will end: an explicit account of what it might have gotten wrong.

A note on the numbers before we start. Every figure below is pulled from our own production records for the month - the daily metrics tables, the per-fill edge ledger, the infrastructure telemetry, and the hedge marks. Where a number is noisy or a series is incomplete, we say so in the text rather than smoothing it into a cleaner story. That discipline is the product; the analysis is downstream of it.

The month in one breath

June's worst six days took SOL down 27% and BTC 19% - the sharpest drawdown of the year. By the close of the month our toxic-fill rate had halved, the spread we captured had doubled off its crash-week low, our capacity never tightened once, and the engine finished green every single day of it.

That sentence is the whole report. Everything below is the evidence for it, layer by layer - and, at the end, the case against it. Here is the month at a glance, in the format every issue of this series will carry, so that a year from now these reads stack into something you can compare in ten seconds:

1 · The price layer

This is the layer you can get anywhere, so we will be brief. June was a two-act month.

Act one, the crash. SOL opened the month near $82 and fell, hard and broadly, to an intraday low of about $60 on the 6th - a 27% drawdown top-to-bottom in six sessions. Bitcoin ran a milder version of the identical shape, from roughly $74,000 to $59,000, a 19% trough. The correlation was the point: this was not a Solana-specific event but a whole-market repricing, everything down together. (Contrast that with the May episode, where SOL fell 12% while BTC barely moved - a single-name move, a different animal entirely. Telling the two apart is most of the job.)

Act two, the grind back. From the 7th onward, SOL clawed its way back to about $72 by month-end - recovering roughly half the fall but still closing the month down around 13% from where it opened. BTC followed, recovering toward $64,000.

If the price layer were the whole story, June's lesson would be the tired one - "hold through the noise" - and we could stop here. We are not going to stop here, because underneath that price, the market was doing something the chart structurally cannot show.

2 · The market-quality layer

Drop one level below price and you stop asking where did it go and start asking how healthy was the market while it went there. Depth, spreads, the cost to transact, how violently quotes were moving - the texture of the market, not its direction. We compress that texture into a single 0-100 reading we call Liquidity Weather, built from depth, volatility, fee environment, inventory pressure, toxicity, and execution quality. It deliberately contains no directional view; a market can be falling and high-quality, or rising and treacherous.

Through the back half of June - the stretch for which we have clean, continuous index readings - the composite climbed from the low-60s, "caution," up to 80, "excellent" - its single best reading of the quarter, with intraday peaks at 87. Spreads corroborate it from another angle: our recorded quoted spreads widened to 2.8 bps on the crash day and then tightened to a month-low 2.3 bps by the 17th-18th, even as the price sat double-digits below where the month opened.

But a composite index hides as much as it reveals, so here is the part a competent read owes you: which components actually moved. When we decompose the Weather recovery from mid-month to late-month, it is not a uniform tide lifting everything.

That -4 on depth is the kind of detail we refuse to round away. The market that emerged from the crash was thinner than the one that went in, but far more rewarding and far less toxic to provide liquidity into. Those are different things, and conflating them is how recaps mislead. The honest one-line summary of the market-quality layer is not "liquidity came back." It is: the market got more profitable and less hostile to trade, on slightly less depth. That distinction will matter in section seven.

3 · The engine layer

Here is the part only we can show you, and it is where June stops being a market report and becomes ours. The engine layer is the market as experienced by capital that is actually in the water - and in a crash, the water bites.

3.1 · Fill quality: were we being picked off?

Every fill we make carries a quality signal called mark-out: after we trade, which way does the mid-price move over the next five minutes? If it consistently moves against the side we just took, we are being adversely selected - "picked off" - which is the defining signature of a stressed, toxic market where the informed traders are running over the liquidity providers. Negative mark-out is the market telling you that you are the slow money.

Watch what our five-minute mark-out did across June:

Alongside it, the toxic-fill rate - the share of our fills whose mark-out breached our adverse-selection threshold - fell from 25.2% in the crash week to 12.9% by the final week. One fill in four was toxic at the bottom; barely one in eight by month-end. And the realized spread we actually captured per fill - the half-spread net of that adverse move - went from a thin, dangerous 0.69 bps in the panic back to roughly 1.0-1.4 bps as conditions normalized. (We will flag one caveat now and bank it for section seven: our fifteen-minute mark-out series is far noisier than the five-minute and we do not lean on it; the five-minute is the signal we trust.)

This is the cleanest statement of the month's thesis, and it is one no external observer could make: the market stopped picking us off, in our own fills, days before the price chart looked recovered.

We noticed the market stop exploiting us well before it stopped frightening everyone else.

3.2 · Execution and infrastructure under load

Fill quality is what the market did to us; execution telemetry is how well the machine itself held up while it happened. Crashes are where engines break - latency blows out, orders pile up, the plumbing backs up exactly when it matters most. Ours, by and large, did not.

The single sharpest stress signature in the whole month is the order-rejection rate. On a normal June day it ran between 0.1% and 1%. On the crash day, the 6th, it spiked to 10.4% - more than one trade in ten bouncing as conditions moved faster than they could clear - before collapsing back to near 0.3% within days. That spike is the engine feeling the crash in real time, and its rapid normalization is the engine feeling the recovery. Slippage and market-impact costs rose with it and receded with it.

What did not break is as important as what did. Execution latency held steady at roughly 15-18 ms through the entire month, crash included, with the tightest internal percentiles barely moving - no blow-out, no backlog spiral. The infrastructure layer absorbed a 27% drawdown without losing its footing. For a strategy whose entire premise is reliable participation, "the machine stayed fast while the market panicked" is not a footnote; it is the foundation the rest of the numbers stand on.

Activity contracted sensibly into the stress rather than chasing it: daily volume fell to its monthly low on the crash day (notional roughly a third below a normal session) and trade counts thinned, which is the correct reflex - quote less, and wider, when adverse selection is peaking, exactly when section 3.1 shows toxicity was worst.

Markets recover in layers. Price, reliably, is the last one to get the memo.

3.3 · Capacity and the size of the book

A question that matters more every month we grow: did the crash squeeze us against our own size limits? No. Capacity utilisation held essentially flat at ~45-47% across the whole month - we were running at well under half of our modelled frontier, with headroom near 54% the entire way through, including the worst of it. We were never forced to trade worse because we were too big for the conditions.

Quietly, and against the grain of the price, assets under management grew through the drawdown - from roughly 83,950 SOL at the start of the month to 88,209 by the 20th, about +5%. Capital came in while the chart was red. We mention it not as a victory lap but because it is the variable we will be watching hardest: stable capacity at today's size tells you nothing about capacity at triple the size, and we would rather flag that now than be surprised by it later.

3.4 · Where the return actually came from

The book was green every single day of June, the crash week included - the market-neutral result we walked through in a separate note. But "green" is a conclusion; the useful question is from what. Decomposing the month's profit by source:

That split - 51% market-making, 49% fee capture, the rebalancers a rounding error - is the quantitative proof of the thing we always claim: the return is manufactured from market activity, and a violent month is a high-activity month. Spread by pair was diversified across the book - SOL/USDC did the heavy lifting (629 SOL) but ETH and BTC pairs contributed meaningfully (WETH 231, WBTC 220, cbBTC 182) - which is itself a quiet step away from being a single-token strategy.

There is a deep point buried in this layer, and it is the one to take away from the whole engine section. To a directional investor, June was a 27% hole and a slow climb out - frightening. To a liquidity provider, June got easier and more profitable to trade almost every day after the first week. Same month, same asset, opposite experience - and which one you live depends entirely on whether your return comes from the price going up or from the market simply being active.

4 · The protection layer

One more layer that only we can show, and we include it precisely because it cost us money - reporting our own bills is the whole point. Our continuous hedge marks begin on the 7th, just after the bottom, so we cannot show you the protection paying into the fall this month. What we can show is the mirror image: across the recovery, as price climbed back from $62 toward $72, the hedge pool's NAV per unit eased from about 0.99 down to roughly 0.90, touching the high-0.78s mid-month. That is the protection giving back premium on the bounce - the exact carry you pay in a recovery, the necessary cost of the payout you would have collected had the fall continued. It is insurance behaving like insurance, and we have written at length about why that carry is not a leak but the price of a floor. The protection layer did not have a triumphant month. It had a correct one.

5 · June in context

A single month means little without the run it sits in, so here is the year around it. June was, by a clear margin, the hardest month to trade in all of 2026 - and that difficulty had been building for two quarters.

The same compression shows up everywhere in the year's record: our win rate has drifted from 91% in January to 77% in June, profit factor from 21x down to 9x, and traded volume has thinned month over month. None of this is alarming - a 9x profit factor and a 77% win rate are still strong numbers, and a market-neutral book is supposed to earn less as edges get arbitraged and competition arrives. But it is the honest frame for June: this was not a healthy market briefly disrupted by a crash. It was an already-tightening market that took a 27% shock and, remarkably, came out the other side with its quality metrics at the best levels of the quarter. The recovery is more impressive read against that backdrop, not less.

6 · What we learned

Three things, stated plainly.

First, liquidity healed before sentiment did. The engine felt the crash in real time - adverse mark-out, one-in-four toxic fills, a ten-percent rejection spike - and then watched the market repair itself in its own fill data days before the price chart looked recovered. If you only watched price, you missed the recovery until it was obvious. If you watched the plumbing, you saw it early.

Second, "healed" was specific, not general. The recovery was a fee-environment and toxicity story, not a depth story - the market got more rewarding and less hostile on slightly thinner books. A lazy read calls that "liquidity came back." The precise read is more useful and more honest.

Third, in a crash, our edge is structural, not directional. Half spread capture, half fee capture, zero from guessing - and the worse the month looked on a price chart, the more raw material the engine had to work with.

For a directional trader, June was a question about price. For us, it was a question about market quality. Price was still damaged at month-end. Market quality wasn't.

When price and market quality disagree, we pay far more attention to the second.

7 · What surprised us

The recovery itself wasn't the surprise - markets bounce, and a violent month is a busy one for a liquidity engine. The surprise was in the composition of the recovery.

Market depth never came back with the rest of it. Every other quality signal we track healed to the best of the quarter - the fee environment doubled, toxicity fell by half, inventory pressure eased - while the order books that re-formed after the crash stayed measurably thinner than the ones that went in. We expected those to move together. They didn't. The market emerged from June simultaneously healthier and shallower: more rewarding to provide liquidity into, and more fragile if it gets hit again.

We are still sitting with what that means. A market that pays you more to stand in it while quietly giving you less room to stand is not an unambiguously good market - it is a more efficient one and a more brittle one at once. If we had to name the single thing June taught us that we didn't already believe, it's that "liquidity recovered" can be true and misleading in the same breath, and that depth deserves its own line in this report every month from here.

8 · What this report could be wrong about

Every issue of this series will carry this section, and it is not boilerplate. A market read with no stated way to be wrong is just confidence wearing a costume. Here is where June's read is most vulnerable.

• The healing might be partly us, not the market. Section 3.2 shows we quoted less and wider into the stress. If our own retreat is what improved our mark-out and toxicity, then some of the "market healed" story is really "we stepped back," and the two are genuinely hard to separate in our own fills. We judge it mostly market - spreads and the Weather fee component moved independently of our activity - but we cannot prove the split cleanly, and we will not pretend we can.
• The structure series misses the worst of it. Our clean, continuous Weather readings run from mid-month; the crash-week structure is reconstructed from engine telemetry and price, not from the index itself. The single ugliest hours of June are inferred, not directly logged. We are fixing the coverage gap; until then, treat the crash-week structure read as the weaker one.
• Depth slipping undercuts the clean narrative. The most bullish framing - "the market fully recovered" - is contradicted by our own depth sub-score, which fell. A thinner market is a more fragile one. If July brings a second shock, the lower depth is exactly where it would hurt, and this month's optimism would look premature.
• One month is one data point. A single heal-before-price episode proves nothing about the next drawdown; correlation between our quality metrics and forward conditions is unestablished over a sample this small. The value of this series is the run of it, including the issues where this read turns out wrong - and we will mark those explicitly when they come.
• Capacity comfort is size-specific. Headroom near 54% at ~84-88k SOL says nothing about headroom at three times the book. As AUM grows, that is the first number we expect to come under pressure, and we would rather pre-commit to flagging it than quietly let it erode.
• The attribution is realized, not risk-adjusted. The 50/50 market-making/fee split is where the profit landed; it is not a statement about which engine ran more risk to earn it. A fuller version of this report will carry risk-adjusted attribution; this first one does not.

Next month, the same seven layers - price, market quality, engine, protection, context - and an explicit, itemized mark against whatever this issue got wrong. A market report is only worth reading if the desk writing it can be caught, and now you know exactly where to try.

Definitions

For readers new to the telemetry, the terms used above, briefly and without hand-waving:

• Mark-out (5-minute): the change in mid-price over the five minutes after one of our fills, signed so that negative means the market moved against the side we took. The standard measure of adverse selection. Persistently negative mark-out means informed flow is trading through us.
• Toxic fill: a fill whose mark-out breaches a fixed adverse-selection threshold. The toxic-fill rate is the share of our fills that qualify - a direct read on how often we were on the wrong side of better-informed traders.
• Realized spread: the half-spread we actually keep after netting out the adverse post-fill move. Quoted spread is what's advertised; realized spread is what survives contact with the market.
• Liquidity Weather: our composite 0-100 market-quality index, built from depth, volatility, fee environment, inventory, toxicity, and execution sub-scores. Direction-free by construction. Detailed in its own explainer.
• Capacity utilisation: the fraction of our modelled capacity frontier we are currently running at. The frontier is the size past which our own trading would move prices against us enough to erode the edge; utilisation below 100% means headroom remains.
• P&L attribution: the decomposition of profit by the engine that earned it - market-making (spread capture), fee capture, and the two rebalancing engines - measured on realized results for the period.