The Regime Flipped on Monday. We Kept Selling Until Friday.

Last week we published ten short case studies and eight of them were winners. This week the same desk sold six times and lost six times. Nothing broke. The tape changed over a weekend, and a system that reads each session from scratch spent four days finding out how thoroughly. The window closed at minus 4.95R across seven trades, a 14.3% win rate, and 9,900 dollars off a simulated 100,000 dollar account at 2% risk. That is the second worst week of the year and the worst since March. Against it, the year to date ledger stands at +28.75R across 220 trades, which puts the same simulated account at $157,519.50. Those two numbers belong in the same paragraph on purpose. A week like this one is what the year looks like from the inside.

Monday: four shorts in sixty-six minutes

The first trade filled at 14:17 UTC and the fourth at 15:23. EURUSD short, then NAS100 short, then US500 short, then GBPUSD short. Four instruments, four separate gates, one posture.

By 16:42 all four had stopped. The equity index shorts were run over from below, NAS100 from 28,970 up through 29,095 and the US500 from 7,604 up through 7,627. The currency shorts failed in the mirror image, EURUSD from 1.15346 up to 1.1557 and Cable from 1.34806 up to 1.34955. Equities bid and the dollar offered is not two failures. It is one risk-on session hitting four positions that were all, in different clothing, short risk.

That is the honest read of the day. The desk did not take four ideas on Monday. It took one idea four times and sized each of them as though the other three did not exist.

Thursday: the only trade that went the other way

On Thursday the gate produced a long. NAS100 at 29,395 against a stop at 29,295, out at the second target of 29,580 the following afternoon, +1.05R on the conservative baseline this recap uses and +1.85R to the furthest target it reached.

It is worth being precise about why that happened, because the tempting story is wrong. The system did not change its mind. It holds no directional view between sessions and cannot be stubborn, because it does not remember Monday on Thursday. It scored a long because the inputs in front of it on Thursday morning scored long: yields retreating below their five day average after failing at 5.025 percent, VIX at 15.73 and falling, both agents bullish. We wrote that one up in full as this week's case study.

Friday: the same axis, the same answer

Friday it sold Cable at 1.33483 and the euro at 1.1464, twenty-one minutes apart. Both are the dollar trade wearing two tickers. Both stopped, the first in twenty-four minutes.

So the week ends where it started, on the wrong side of the same axis, with one Thursday in the middle where the tape happened to hand the gate a long instead.

Pattern of the week

The pattern this week is not in what we entered. It is in how fast the entries died.

Six losing shorts, average time from fill to stop of sixty-seven minutes. Four of the six were finished inside an hour. The US500 lasted twenty-two minutes. The winner, by contrast, ran for twenty-three hours and sixteen minutes.

That gap says something specific. A trade that is merely early tends to go against you, come back, and take hours to resolve. A trade that is on the wrong side of the prevailing flow does not get that treatment. It gets run over immediately. Six immediate stop-outs in one week is not six pieces of bad luck arriving separately, it is one piece of information arriving six times.

What it is not is a discipline failure, and we want to be exact about that because it would be the comfortable conclusion. Every one of these entries waited for its trigger. The US500 short sat through eight consecutive evaluations that all declined to enter, each one noting that price was inside the zone but the rejection candle had not printed. On the eighth it printed, cleanly, at 15:10 UTC. The entry was correct by every rule we have written down and the position was underwater within minutes.

Decision highlights

The US500 short is the decision worth defending even though it lost. Eight consecutive evaluations declined to enter, each one stating that price was inside the 7,604 to 7,608 zone but that no completed five minute candle had closed back below the level with a bearish body. The ninth read printed exactly that rejection at 15:10 UTC and the position went on. Twenty-two minutes later it was stopped. Nothing in that sequence is a process failure, and if we retuned the system to avoid this trade we would be retuning it to avoid a rule that has paid for itself all year.

Monday's sizing is the decision we would take back. Four positions opened inside sixty-six minutes, each cleared on its own merits, each sized as a standalone 1R risk, and all four short risk in a session that turned out to be risk-on. The gate is per instrument and it has no view on how much of the same bet is already on the book, which is exactly how a minus 4R session gets assembled out of four individually reasonable decisions.

Friday's second entry is the same decision in miniature. The GBPUSD short had already stopped at 14:51 when the EURUSD short filled at 14:47, so the two overlapped rather than sequenced, and the dollar was the common factor in both. One of those trades was a position and the second was the same position again.

Loss worth learning from

Where the 4.95R went

Monday accounts for four of the six losses and 4R of the total. The remaining 0.95R of damage is Friday's two shorts netted against Thursday's win.

The individual entries do not teardown well, and that is the finding rather than an evasion. There is no mistimed entry to point at, no stop placed on the wrong side of a level, no trade that skipped its confirmation. Entry prices were inside their zones. Stops sat beyond structure. Each loss was exactly 1R, which is the system doing precisely what it was built to do when it is wrong.

The teardown that does produce something is the one level up. Four correlated shorts on Monday and two more on Friday is six expressions of one macro posture across a week in which that posture was wrong from the opening bell. Had those six been sized as what they actually were, the same six calls would have cost a fraction of 4.95R.

From the desk

From the desk

Minus 4.95R is 9,900 dollars off the simulated account and it takes the equity line from 100,000 to 90,100 on the week. We would rather publish that number in the same format we publish the good ones than quietly change the format.

The year to date ledger reads +28.75R across 220 trades at a 56.82% win rate. Run through a static 100,000 dollar account at 2% risk that is $157,519.50. Run through the same account compounding, where each position is sized off the balance it actually has rather than the balance it started with, it is $169,638.84. The gap between those two figures, a little over 12,000 dollars, is not extra edge. It is the same R banked in a different order of operations, and it only exists because no single week has been large enough to break the sizing. A week like this one, which is the second worst of the year, moves the static line by 9,900 dollars and the compounded line by a comparable amount, and neither of them threatens the account. That is the entire argument for fixed fractional risk, and it is easier to make honestly in a losing week than a winning one.

Two pieces of housekeeping on the numbers. The year to date figure is sealed at the August close by design, so it does not yet include September. September itself is open and sits at +3.92R through Friday, down from a peak of +8.87R on the tenth, with this week accounting for most of the round trip.

Next week the desk will do what it did this week, which is read each session from scratch and take what clears. If the risk-on tape holds, the long setups will keep clearing and we will keep taking them. If it does not, the shorts come back. We do not get a vote and neither does the model.

What we're tuning

We are not touching the entry gate. Six losses where every entry satisfied its trigger is not evidence that the trigger is too loose, and the fastest way to wreck a system that is +28.75R on the year is to retune it against the week it just had.

The correlation question is different, and it is the one this week actually raised. The gate evaluates each instrument in isolation, so four short-risk positions opening inside sixty-six minutes reads to the system as four independent decisions when the tape treats them as one. We are looking at whether portfolio-level exposure should throttle the fourth correlated entry in a session rather than the first. That is a real change with real costs, because the same throttle would have capped the ten short week that preceded this one, and we would rather measure it properly than ship it because we just had a bad Monday.