7-0: A Record Week for SkyAnalyst, and the Month It Rescued

Seven trades. Seven winners. No stops. The week of August 24 closed at +8.02R with a 100% hit rate, the best week SkyAnalyst has produced in 2026, and it landed at the end of the worst month the desk has had all year. That combination is the whole story, and we want to be careful about how we tell it. A 7-0 week is not evidence that anything has been solved. It is five sessions. What makes it worth writing up is not the streak, it is that the week followed directly from two pieces of research we published days earlier and a change we shipped on Thursday because of them. Through Aug 31, 2026, the system stands at +31.43R since its Jan 12 inception across 176 trades at a 59.09% hit rate. A $100,000 simulated account at 2% risk per trade sits at $162,868.55. This week moved that account from $100,000 to $116,055.91 on the window alone, a simulated +$16,040 in five sessions.

Act one: the month that forced the question

August was, until Monday, the worst month of the year. Thirty-six trades through the 23rd for minus 9.70R, and three consecutive losing weeks. The losses were not dramatic individually. Every single one closed at exactly minus 1R, which is the system working as designed. The problem was the count.

When a book loses that consistently for three weeks, there are two honest explanations. Either the edge has decayed, or the book contains something that is dragging on the rest of it. We spent the month testing the second explanation rather than assuming the first.

Act two: measuring the two halves apart

The answer, when we finally cut the data properly, was uncomfortable. Every short book on the roster beat its random-walk baseline. Almost every long book did not. We published that in full, including the working, in the shorts-versus-longs study, and we published the stop-width companion piece beside it.

The finding on its own changed nothing, because until Thursday a trader was one book. EURUSD was EURUSD, longs and shorts summed into a single line, which meant a profitable short book and a leaking long book cancelled each other out and reported as mediocre. You cannot switch off half of a trader you only measure as a whole.

Act three: the split, and what followed it

On Thursday we separated every AI trader into an independent long book and an independent short book, each with its own continuous record. That is the change. It sounds administrative and it is not, because it turns a measurement into a control: a book that cannot clear its baseline can now be switched off without touching the half of the same instrument that works.

The first two trades taken after the split went live were NAS100 long and US30 long, the two long books the research had said to keep. Both won. That is not proof of anything, and five sessions never will be, but it is the first week where the roster we ran was the roster the evidence pointed at.

Pattern of the week

Every winner was a pullback, and not one was a chase

The cleanest pattern in the week is that the desk did not take a single breakout. All seven entries were retracements into a level: a VWAP rejection on US500, a failed reclaim on US30, a broken-support retest on Cable, a Fibonacci retracement on the euro, a VWAP bounce on the Nasdaq, a retest long on the Dow, a continuation pullback on USDCAD.

That is not a setting we changed this week. It is what the confluence gate has always preferred, and it shows up most clearly on weeks when the tape actually offers the pullback. Three of the seven write-ups contain an explicit refusal to chase, written into the analysis before price arrived: a named zone with an instruction to stand aside if price broke away without returning.

The honest caveat is that a week where every retracement holds is a week that flatters this style. The same seven setups in a trending tape that never pulls back produce zero entries, not seven winners.

Decision highlights

The desk refused a price it liked on Monday. The US500 analysis named a short zone of 7664 to 7668 while price was still eight points below it, and wrote down that a chase entry near 7655 would deliver 0.7R against a 1.5R minimum, so the correct action was no trade. Price came back to the zone and the trade filled at 7664. The decision worth noting is not the entry, it is that the refusal was priced before the opportunity existed.

On Tuesday the Dow setup scored 89% on its first evaluation and the system declined it, then declined ten more, and entered on a reading of 61%. Confidence never was the gate; the three trigger conditions were. A reader assuming the system enters when it is most certain would have expected that trade thirteen minutes and roughly fifty points earlier.

Thursday's US30 long is the one we will remember, because it is the trade that tested the week's own thesis. We had published a correction that morning arguing the US30 long book should stay switched on, against an earlier reading that said retire it. Hours later that book took a retest long and closed +0.95R. One trade does not vindicate a decision, but it was a pointed way to start.

From the desk

From the desk

The number we would rather you take from this week is not the 7-0. It is that the month is still negative. August finished at minus 1.67R across 43 trades, 21 winners against 22 losses, and one exceptional week did not make it a good month. It made it a nearly-flat one.

Through Aug 31, 2026, a $100,000 simulated account at 2% risk sits at $162,868.55 on static sizing. Compounding the same sequence of R through the same 2% rule produces $180,579.40. The gap between those two figures, roughly $17,700, is not extra edge. It is the same +31.43R compounding on a growing base, and it only exists because no single loss was ever allowed past one unit of risk. A book that takes one outsized loss loses that gap permanently.

We spent August measuring rather than reacting. The research said the long side was the leak, the split made the long side switchable, and the first week under the new arrangement went 7-0. Any one of those sentences on its own would be a weak claim. The sequence is the point, and the sequence is checkable, because we published each step before we knew how the week would end.

What we're tuning

The direction split is the change, and it is now a setting rather than a rewrite. Each trader is an independent long book and short book with its own continuous record, so a book that stops clearing its baseline can be switched off without disturbing the other half of the same instrument. On the separated numbers that means NAS100 and US30 longs stay on, and the GBPUSD, EURUSD and USDCAD long books are the ones under review.

We also brought USDCAD into the published standings this week. Both books have been running live since July and were the only active traders sitting outside the record, which meant the roster we published and the roster we ran were not the same thing. That is fixed, and the effect on the year is about +0.06R, so nothing in the historical numbers moves.