July 2026

The number is -9.96% net. It was not produced by a regime the system failed to read, or by a broad deterioration in the book. 217 of the month's 218 closed trades behaved normally and returned +2.61% between them. One position lost 11.03% of account equity.

This letter is mostly about that one position.

01 / Performance

The strategy returned -9.96% net across 25 active trading days. 23 of those days closed positive.

On June 30 at 16:45 UTC the system opened a long in INUSDT. It closed on July 2 at 10:52, one day and eighteen hours later, at an average exit price 55.75% below the average entry. The realized loss was 11.03% of account equity in a single trade. Maximum adverse excursion during the hold reached -74.6%. The position never traded favorably at any point after entry.

Everything else in July was ordinary. Excluding that position, the remaining 217 closed trades produced +2.61% across 34 symbols with two small losses, which would have made July a positive month. I mention that arithmetic not to discount the loss but to locate it: this was a single-position event, not a strategy-wide one.

All returns are reported net of commissions and funding costs.

02 / Risk & Execution

Maximum drawdown was 12.41%, recorded on July 2. That is now the strategy's largest realized drawdown since inception, and it replaces the -7.8% figure previously carried across our materials. Every document has been updated.

What happened to this position is worth setting out precisely.

INUSDT satisfied every entry, volatility, and risk filter at the moment of execution. Those filters exist specifically to keep the system out of abnormal conditions, and on the information available at entry the setup was valid. Within moments of execution the token moved into what we assess to have been a coordinated pump-and-dump.

That distinction matters because of what the strategy actually trades. Our edge is mean reversion against forced-flow dislocations: leveraged holders being compelled out of positions at prices they would not choose, followed by a snap back. A manipulated distribution looks similar at the point of entry and behaves nothing like it afterwards. There was no forced cohort to revert against, because the move was not caused by one. The premise the position was opened on had been invalidated by the time it was filled.

The position was closed by discretionary review rather than by an automated trigger. Having concluded that the trade no longer reflected the statistical conditions the strategy requires, we closed it to stop further loss rather than continue holding for a reversion that had no mechanism behind it. Forty-two hours elapsed between entry and that decision, against a median hold of eighteen seconds.

Two things follow from that, and I will not soften either. First, our prior description of this control referred only to automated volatility stops; it did not account for discretionary closure of an instrument that falls outside the model's assumptions. We have updated that description across our materials to reflect how the control actually operates.

Second, the loss crossed a month boundary. The position opened on the final afternoon of June and closed on July 2, so June's reported +7.22% was accurate on a realized basis while the book already carried this position deeply underwater. We report realized results and have applied that convention consistently, but June's letter reported a 0.55% maximum drawdown without disclosing an open position in that state, and a note has been added to it.

On capital: the loss consumed roughly 60% of the profit accumulated over the prior quarter. Client principal was not touched, and the account finished July above its starting capital.

03 / Market Structure

The market context matters less this month than the position does, but it should be recorded.

July was quiet. The late-June liquidation cascade had already cleared leverage out of the system, and average daily liquidations across crypto derivatives ran well below the $400 million to $500 million range typical for the year. Bitcoin recovered roughly 6% off its early-July low to finish near $63,900, and the CoinDesk 20 posted its largest monthly advance in a year, but the advance came on thin participation rather than forced transacting. US spot Bitcoin ETFs drew approximately $205 million in net inflows, the smallest monthly total since the products launched.

Genuine forced-flow opportunity was scarce, which is why activity outside the INUSDT position fell to roughly 40% of June's trade count. In a month with little to trade, a single bad position is not diluted by anything.

04 / Looking Ahead

The work is on token selection and market integrity screening rather than on entry timing or sizing. The filters did what they were specified to do; the specification did not contemplate an instrument being actively manipulated while it passed them. Detecting that condition before committing capital, rather than reacting to it afterwards, is where the framework is being extended.

I would rather report this month plainly than frame it. Nineteen months of results now include one that lost close to ten percent, the maximum drawdown across the record is 12.41% rather than 7.8%, and both figures are what an allocator should underwrite against.

The strategy's other parameters held. That is worth something, but it is not the point of this letter.