September 2026: Thirty-Two Shorts, One Thesis, and Two Bad Sessions

September was the month the desk ran one thesis, and learned what that costs. The headline is +9.59R across 37 trades, 23 winners against 14 losers, a 62.2% win rate. Underneath it sits a remarkably narrow book. Thirty-two of those entries were shorts: sell the rally into VWAP, sell the retest of a broken low, sell the failed reclaim. The same shape showed up on US30, NAS100, US500, EURUSD, GBPUSD and USDCAD. When the tape agreed, it paid on several instruments in the same session. When it did not, it stopped out several instruments in the same session. The month closed on the right side of that. The final two trades, a NAS100 long off breakout support on September 30 and a US30 short on a failed retest under VWAP the same afternoon, both paid and took the month to its high. Through Oct 1, 2026, the system has banked +38.35R YTD since the Jan 12 inception. A $100,000 simulated account risking 2% per trade sits at $176,703 on static sizing.

Act I: The read that paid

The first two weeks of September belonged to the short book. September 1 opened with three winners in under an hour, including the month's best trade, a US500 short into VWAP and the prior day's low that paid +2.14R on the TP1 baseline. September 10 did it again with four winners in one session across GBPUSD, NAS100, US500 and US30.

By the close on the tenth the month stood at +8.87R. It was not a clean climb. September 3 and 4 had already cost three losses in a row, one of them the month's only losing long, and the curve had dipped to +3.44R before the second week put it back. But the direction was right and the desk kept expressing it.

Act II: The same bet, four times

September 14 is the session the month will be remembered for. Between 14:17 and 15:23 UTC the desk sold EURUSD, NAS100, US500 and GBPUSD, and every one of them stopped out. Four different setups on four different instruments, and one underlying opinion about the dollar and risk appetite, which the tape had stopped sharing.

The week that followed added two more short losses on September 18 and one NAS100 long winner in between. The month went from +8.87R on the tenth to +3.92R on the eighteenth. We covered that week in detail in the September 14 weekly recap, and its conclusion still holds: the entries waited for their triggers, and the triggers fired into a regime that had already turned.

Act III: Six in a row, then three back

The dollar trend resumed on September 21 and the book found it immediately. Six straight winners across GBPUSD, NAS100 and US30 carried the month to +9.39R by the middle of the 23rd.

Then September 24 repeated September 14 on a smaller scale. NAS100, EURUSD and GBPUSD shorts, all within about forty minutes, all stopped. The pattern was the same, the location was late in a move that had already been made, and the book gave back 3R in a session. The last week of the month was three trades, three winners, and a close at +9.59R, the high of the month.

Pattern of the week

The pattern of September was concentration, not a setup.

On a per-trade basis the desk did what it is built to do. Every entry waited for its trigger, every loss was exactly -1R, and the setups were the ordinary pullback-sell shapes that the desk trades in any trending tape. What was unusual was how many of them were the same trade underneath. A short on US500, a short on NAS100 and a short on EURUSD taken inside the same hour are three setups on the chart and one opinion about the market.

That is why the month's losses arrived in clusters. Thirteen of the 14 losses were shorts, and seven of them fell on two days. On the days the thesis held, the concentration showed up as four winners in a session. On the days it did not, it showed up as four stops. The net was strongly positive, and we want to be clear that this is not a complaint about the result. It is a description of the risk the result was built on.

Decision highlights

The desk held risk at exactly 2% per trade through both of the month's bad sessions. Every one of the 14 losses in the table is -1R and -$2,000 on the simulated account, including the four on September 14 and the three on September 24. There was no attempt to size down while losing or size up during the six-trade run, and that flat line is why the month's worst stretch cost 4.95R rather than more.

NAS100 was allowed to go long while the rest of the book was short. Five times in September its evaluations cleared a long pullback against the grain of the desk, and four of those paid. A system that forced every trader into the same direction would have sold those bounces instead, and the NAS100 line would look like the FX lines.

After each of the two cluster days, the desk went quiet. There were no fills on September 15 or 16, and none on September 25. We did not write a rule that pauses trading after a bad session, and these gaps are not one. They are what the next evaluations returned once the regime had turned, which is the behavior we want from a desk that reads each session from scratch.

Loss worth learning from

The loss worth learning from is not one trade. It is September 14, because the four stops that day are more instructive together than any of them is alone.

Every one of the four was graded C+, like most of the month. Every one waited for its trigger: the US500 short, as the weekly recap documented, sat through eight consecutive evaluations that declined to enter before the rejection candle finally printed. Each trade, reviewed on its own, was a correct application of the rules we have written down.

The problem is visible only when you stack them. EURUSD at 14:17 UTC, NAS100 at 14:42, US500 at 15:16, GBPUSD at 15:23. Four tickets, one thesis: the dollar stays bid and risk stays offered. When that thesis turned on Monday, the desk did not have four independent chances to be right. It had one chance, taken four times, and it cost 4R in a single session.

That is the part of September we most want readers evaluating the system to see. The per-trade discipline held. What the per-trade rules do not yet see is the book as a whole.

From the desk

September added +9.59R and $19,184 to the simulated account.

Through Oct 1, 2026, the system stands at +38.35R across 257 trades since the January 12 inception, 148 winners against 109 losers, a 57.59% win rate. On a $100,000 account at 2% risk, static sizing puts that at $176,703, up from $157,519 at the end of August. Compounding the identical sequence of trades, where each position is sized off the balance the previous one produced, puts it at $203,855. The gap between those two figures is $27,152, and it is not the product of a better trade. It is the same R, sized with discipline and allowed to work on itself. It only holds because no single session has been large enough to break the sizing, and September's two worst sessions, 4R and 3R, did not come close.

That is the case for fixed fractional risk in one month. The desk took a four-stop Monday and a three-stop Thursday and the next position was sized exactly the way the first one was. There was an account left to take the six-trade run that followed, and the month finished at its high.

October opens with the same question the month closed on. If the dollar trend keeps running, the same books will keep finding the same trade, and we will keep publishing the days it fails in the same format as the days it pays. If the trend is done, the plans will say so, and the book will look different. We do not get a vote, and neither does the model.

What we're tuning

The open question September hands us is correlated exposure inside a session. Today each instrument trader evaluates its own chart and its own plan, and the Risk Agent sizes each trade on its own merits. Nothing yet asks whether the third short of the hour is really a new trade or the same bet sized up. We are examining what a same-session concentration check would have done to September, and we will not ship one until we can show it would have cut the cluster days without cutting the four-winner sessions that paid for them. We have not changed anything yet, and we will say so plainly when we do.