The Week Our Own Selection Gate Refused to Pick a Loss
This report normally tears down the best-graded loss of the week. This week it could not: neither loss graded above C+, because almost nothing did. Two stop-out
Before the trade, meet the system.
SkyAnalyst is not one AI trader. It is four specialist agents, each with its own data pipeline, each maintaining state between evaluations, and each required to agree before a position is sized. They don’t chat in prose. They write structured messages to a shared state object that each reads on every evaluation cycle.
This report has a rule about which loss it tears down. It picks the best-graded one, because a well-formed trade that failed anyway teaches more than a poorly-formed one that deserved to. This week the rule refused to run. It requires a loss graded B or better and there was not one, so the selection threw an error rather than returning a trade. We could have quietly overridden it and written the usual article. Instead the refusal is the article, because the reason behind it describes the week better than either loss does: of ten trades taken, eight graded C+, and that includes both losses and six of the eight winners. Through Sep 14, 2026, the desk stands at +28.75R across 220 trades since the January 12 inception at a 56.82 percent strike rate. A $100,000 account at 2% risk sits at $157,519 on static sizing, having been 1.91 percent below its peak at the worst point of this week.
Act I: Tuesday, USDCAD
A short entered on Tuesday afternoon and stopped out for exactly 1R. Grade C+.
The setup was a bearish continuation pullback in a week where every cross-asset input supported shorts, and it did not work. There is no deeper failure to report: the entry cleared its gate, the Risk Agent sized it against structure, price went the other way, and the position closed at the level that was set before it opened.
It is the book's second September loss from two trades, which is worth noting and not worth acting on. Two trades is not a sample and the alternative, pulling an allocation after two stop-outs, is how a desk ends up with no books left.
Act II: Wednesday, US500
A short entered on Wednesday afternoon and stopped out for exactly 1R. Grade C+.
The S&P book took three trades this week and this was the one that failed. The other two returned +1.36R and +0.67R, which makes this the middle case rather than an outlier: same instrument, same direction, same week, three different outcomes from setups in the same quality band.
Neither loss clustered with the other. The longest losing streak in the window was one, which at a 56.82 percent strike rate is an unremarkable result over ten trades.
Act III: The refusal
Both losses graded C+, so the selection gate found nothing to pick.
The gate is not broken. It was written on the assumption that a week contains a range of setup qualities and that the instructive failure sits near the top of that range. This week had almost no range: eight of ten entries scored C+ before the outcome was known, the other two scored B and B+, and both of those won.
A tool refusing to produce output because its input assumption does not hold is a tool working. The useful thing is to report why rather than to force it and pretend the selection meant something.
Every trade the system took.
| Date | Time | Instrument | Dir | Model | Setup | Grade | R | $ Sim | Result | Details |
|---|---|---|---|---|---|---|---|---|---|---|
| Sep 8 | 15:50 UTC | USDCAD | Short | USDCAD SHORT (Bearish Continuation Pullback) | C+ | -1.0R(SL) | -$2,000(SL) | Stop hit | - | |
| Sep 9 | 15:50 UTC | US500 | Short | US500 SHORT, Bear Flag Breakdown Continuation (Pullback Entry) | C+ | -1.0R(SL) | -$2,000(SL) | Stop hit | - |
Dollar figures are simulated on a $100,000 account at 2% risk per trade. Actual subscriber P&L varies with account size. Past performance is not a guarantee of future results.
Pattern of the week
The pattern this week is that setup grade and outcome had no relationship at all.
Ten trades: two graded B or better, eight graded C+. Both losses were C+. Six of the eight winners were also C+. Ranking the week's entries by grade before any of them resolved would have told a reader essentially nothing about which would pay.
That is not evidence the grading is broken. Grades are assigned at entry, before the outcome is known, and over a large sample higher grades do resolve better, which is the entire basis on which this report normally selects its teardown. What ten trades cannot do is demonstrate that relationship, and a distribution this narrow cannot demonstrate anything at all.
Compare it to August, where the losses had a clear and actionable pattern: the GBPUSD long book took ten trades in one direction, won two, and gave back 6.18R. That was a book on the wrong side of a regime and it needed switching off. This week's two losses share no instrument, no setup type and no session, and they sit inside a week that finished +5.43R. The correct response is nothing, which is harder to write about than an intervention.
Decision highlights
We overrode the selection gate rather than skipping the report. The gate exists to stop us tearing down a weak loss and calling it instructive, and the honest way to bypass it is to say that we did and explain what its refusal revealed. Silently forcing it and presenting the result as a normal selection would have been the dishonest version of the same action.
We changed nothing in response to either loss. Both were C+ setups that stopped out at exactly 1R, neither repeated the other, and the longest losing streak in the window was one. The standard for acting is a repeated one-sided failure across a meaningful sample, which is what August produced and what nothing here resembles.
We fixed the setup grade tile on the weekly scoreboard this week. It had been printing a fixed value on every recap regardless of what the desk traded, which meant a week of C+ entries was published under the same quality claim as a week of A setups. It now averages the real per-trade grades, which is how this week's C+ became visible at all.
Six instruments, six stories.
EURUSD lost nothing this week because it did not trade. No setup cleared the confluence threshold and the book sat out rather than add a sixth short to an already one-sided position.
All EURUSD this week →GBPUSD lost nothing this week. Its single trade was a winning short that needed nine refusals before the tenth evaluation cleared it.
All GBPUSD this week →US30 lost nothing this week. It took three trades, won all three for +2.58R, and has now won every September trade it has taken.
All US30 this week →NAS100 lost nothing this week. Its two shorts returned +2.02R combined, the second carrying half the stop width of the first.
All NAS100 this week →USDJPY did not trade this week, so it lost nothing. Six trades all year and the book remains deliberately on a short leash.
All USDJPY this week →US500 accounts for one of the two losses, a short on Wednesday for -1R, from three trades that also produced +1.36R and +0.67R. The widest spread of outcomes of any book this week.
All US500 this week →USDCAD accounts for the other loss, a short on Tuesday for -1R. It is the book's second September loss from two trades, which is worth watching and not yet worth acting on.
All USDCAD this week →Loss of the week: US500 Short · -1R
Losses worth learning from
US500 Short, Wednesday, -1R, grade C+
The S&P book took three trades this week and this is the one that failed.
The entry was a bear flag breakdown continuation, taken in the same risk-off compound that produced the week's other nine shorts: yields above their EMA and rising, VIX expanding, breadth deeply negative. It cleared its gate, the Risk Agent sized it against structure, and price went the other way to the stop.
What makes this worth publishing is the company it keeps. The same book, same direction, same week, produced +1.36R on Tuesday and +0.67R on Thursday. Three trades from one instrument in one regime, all in the same C+ grade band, and the outcomes ranged from a clean winner through a marginal one to a full stop-out.
There is no adjustment that separates those three in advance. Anyone claiming otherwise is reading a pattern into three data points, which is the specific error this report is written to avoid.
USDCAD Short, Tuesday, -1R, grade C+
A bearish continuation pullback that stopped out at its predetermined level.
The interesting detail is not the trade, it is the book. USDCAD has now taken two September trades and lost both. That is the kind of fact that invites action: pull the allocation, tighten the threshold, review the parameters.
Two trades is not a sample. The book joined the published standings in August on the back of four trades and three winners, which was also not a sample. It has done nothing across six September and August trades that distinguishes it from ordinary variance in either direction.
The discipline being exercised here is doing nothing while a number looks bad, which is the same discipline as doing nothing while a number looks good. Both are how a system avoids being tuned by its most recent outcome.
On a $100k account at 2.0% risk per trade.
Each trade risks +$2,000 (1R). The system's actual scale-out behavior may differ, see disclaimer.
| Scenario | R-multiple | Profit on $100k |
|---|---|---|
| Window drawdownActual | -2R | −$4,000 |
All six agents combined.
From the desk
From the desk
The losses are the least interesting thing that happened this week, and we are publishing them anyway because the schedule does not bend for a good week.
Two stop-outs, minus 2.00R, both at exactly the designed size, inside a week that finished +5.43R with an 80 percent strike rate. Maximum drawdown 1.91 percent. Longest losing streak one. There is no story of damage here and it would be dishonest to manufacture one.
Through Sep 14, 2026, the desk stands at +28.75R across 220 trades since the January 12 inception, 125 winners against 95 losers. On a $100,000 account at 2% risk, static sizing puts that at $157,519 and compounding the same sequence puts it at $169,638. Ninety-five of those 220 trades cost exactly 1R each. That number is the one most performance pages leave out and it is the one that determines whether anyone can actually hold the position.
What this week did produce is a finding about our own instrumentation. The selection gate refused to pick a teardown because no loss graded above C+, and investigating why surfaced that almost nothing graded above C+, winners included. That in turn surfaced that the scoreboard's setup grade tile had never been computed at all. A report that only ever confirms what the desk already believes is not doing its job, and this one caught something.
What we're tuning
Nothing changes on the back of these two losses. They stopped out at exactly 1R each, they share no instrument, setup type or session, and they sit inside a week that finished +5.43R with a maximum drawdown of 1.91 percent. The threshold for acting on a book is a repeated one-sided failure across a meaningful sample, and two unrelated stop-outs is not that.
One thing did change this week, and it came out of writing this report. The weekly scoreboard's setup grade tile had been publishing a fixed value on every recap regardless of what the desk had actually traded, which meant a week of eight C+ entries carried the same quality claim as a week of A setups. It now averages the real per-trade grades. That is how this week's C+ became visible, and it is why the selection gate's refusal turned out to be describing something real rather than being an inconvenience to override.
What the numbers actually mean
Two losses in a ten-trade week at a 56.82 percent strike rate is close to the centre of the expected distribution, and the more interesting statistic this week is the grade spread underneath the results.
The arithmetic on the losses first. At a 56.82 percent strike rate the expected number of losses in ten trades is about 4.3, so two is a favourable draw rather than a sign of improvement. The probability of two or fewer losses in ten trades at that rate is roughly 9 percent, which means a week this clean should appear about once every eleven weeks. It arrived slightly ahead of schedule and it will be paid back.
Van Tharp's framing of R-multiples is the reason the loss count matters less than it appears to. Every loss on this desk costs exactly 1R by design, so the question is never how many losses occurred but whether the winners are large enough relative to that fixed cost. The current R target sits at 0.93, meaning the average winner needs to return about 0.93R for the system to hold its edge at this strike rate. This week's winners averaged well above that, which is why two losses did not prevent a +5.43R result.
The grade distribution is where a reader should focus. Eight of ten entries graded C+ before the outcome was known, and that band contained both losses and six of the eight winners. Over a large sample setup grade and outcome correlate, which is the assumption this report's own teardown selection is built on. Over ten trades they did not correlate at all. This is the practical content of what Schwager's interview subjects keep describing across three decades of Market Wizards: that the sample size required to distinguish skill from variance is far larger than the sample any individual week provides, and that the discipline of a good operator lies in not updating a process on evidence that cannot support the update. An 80 percent week on C+ setups is a fact about this week. It is not yet a fact about the system.
At a Glance
Drawdown questions
Why publish a losses report in a week with eight winners?
Because the schedule does not bend for good weeks, and a record that reports losses only when they are dramatic is not a record. Two stop-outs at exactly 1R inside a +5.43R week is what the system operating normally looks like, and a reader deciding whether it is usable needs the ordinary weeks as well as the bad ones.
What does it mean that the selection gate refused?
The gate picks the highest-graded loss for teardown, requiring at least a B, on the reasoning that a well-formed failure teaches more than a weak one. Both losses graded C+ so it threw rather than selecting. We overrode it deliberately and made the refusal the subject, because the reason behind it describes the week better than either loss does.
Is a C+ average setup grade a problem?
It is a caution rather than a problem. Grades are assigned at entry before outcomes are known, and eight of ten entries graded C+ including six of the eight winners. The week returned 80 percent on ordinary setups, which is variance running favourable. The risk is reading it as a new baseline and sizing accordingly.
USDCAD has lost both its September trades. Why keep it?
Because two trades is not a sample in either direction. The book joined the published standings in August on four trades and three winners, which was equally not a sample. Nothing across its six recent trades distinguishes it from ordinary variance, and pulling an allocation on two stop-outs is how a system gets tuned by its most recent outcome.
Why did every loss cost exactly 1R?
Because position size is calculated so the distance from entry to stop equals a fixed 2% of the account, whatever the instrument. The system does not try to make losses small, it makes them identical, so the only variable determining profitability is what the winners return relative to that fixed cost.
Trade with the system that publishes its drawdowns.
Subscribers receive every signal, winners and losers alike, three minutes before entry, with full reasoning.
Dollar figures are simulated on a $100,000 account at 2% risk per trade. Drawdown trajectories shown reflect a small window sample size and are not projections of forward performance. Past performance, including losses, is not a guarantee of future results. Actual subscriber P&L varies with account size and execution. YTD context: +28.75R YTD across 220 trades, see stats strip.
From the SkyAnalyst Journal
Eight Winners, Two Losses, and a Week That Graded C+
Ten trades, eight winners, +5.43R and an 80 percent strike rate, the best week of the month. The setup grades tell a quieter story: eight of the ten scored C+ before the outcome was known.
Four Shorts in Forty-Eight Minutes, on One Reading
Between 14:20 and 15:08 the desk sold Cable, the Nasdaq, the S&P and the Dow. One macro read, four instruments, four stops, four winners. This is the Dow leg, and the question is what that correlation costs.
The Winner That Paid Less Than a Loss Would Have Cost
This trade was right and returned +0.67R. The target sat 14 points away and the stop sat 21, so being correct paid two thirds of what being wrong would have cost. It is the least flattering trade of the week.