Three stops, one regime: the week the tape refused to confirm
Three losses, 2.25R given back against a year that still reads +20.43R. The honest portfolio view: what every stop taught us, and what the drawdown curve says a
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.
The week of May 11 closed three trades and stopped out of all three. US30 short on Tuesday, USDJPY long and NAS100 short within ten minutes of each other on Wednesday. Total damage: 2.25R, or roughly 4,500 dollars on a 100,000 dollar simulated account sized at 2 percent risk. No winners reached the books inside the loss-attribution sample, though the equity walk shows Friday recovered the ground and then some. We publish the drawdown weeks for the same reason we publish the green ones. Through May 18, 2026 the system carries plus 20.43R for the year, and a 100,000 dollar account sized at 2 percent sits near 140,857 dollars on a static basis. This week gave back about 4,500 dollars of that figure. Stated as a ratio rather than a headline, the system surrendered roughly 11 percent of one week against a cumulative plus 20.43R since the January 12 inception. That contrast is not spin. It is the reason a single losing week is data, not a verdict. Every stop this week was a setup the rules cleared and the tape declined to confirm. None of them were rule violations. That distinction is the spine of the report.
Tuesday: one stop, clean and isolated
US30 printed a short setup on Tuesday afternoon, May 12, on a sell-the-pullback read into the opening-range high. The Trend Agent scored the structure, the Macro Agent did not veto, and the position sized in. Price did not roll over. It pushed back through the pullback shelf and tagged the structural stop for a clean minus 1R, about 2,000 dollars on the simulated account. The trade was a C-plus setup, acceptable rather than pristine, and it failed the way acceptable setups fail: the level it leaned on did not hold. One stop, isolated, no streak yet. The equity walk shows the dip and the partial recovery into Tuesday's close.
Wednesday: two stops inside one hour
Wednesday, May 13, is where the week's character formed. At 14:33 UTC a USDJPY long triggered on a New York morning VWAP pullback continuation, graded B, the cleanest read of the three. Ten minutes later, at 14:43 UTC, a NAS100 short triggered on a VWAP and Fibonacci rejection. The yen long stopped for a shallow minus 0.25R; the index short stopped for a full minus 1R and became the loss of the week. Two positions, opposite instruments, the same ten-minute window, both stopped. The drawdown curve bottoms here at minus 2.4 percent. This is the act that matters, because the two stops were not independent draws from the same distribution. They shared a tape.
Friday and the recovery
The loss-attribution sample for this report is the three stops; it does not carry the winners. The drawdown curve does, because peak-to-trough math needs the green prints to be honest. By Friday, May 15, the equity walk had climbed back above the Wednesday trough and printed a fresh high near 111,923 dollars on the simulated account. We are not reporting Friday as a victory lap. We are reporting it as the structural answer to the question every drawdown raises: did the system keep its sizing discipline through the red, and re-engage the same way it always does. It did. The recovery is not the story. The behavior through the drawdown is.
Every trade the system took.
| Date | Time | Instrument | Dir | Model | Setup | Grade | R | $ Sim | Result | Details |
|---|---|---|---|---|---|---|---|---|---|---|
| May 12 | 15:42 UTC | US30 | Short | Claude Opus 4.6 | Sell the Pullback into OR High / 5m Resistance | C+ | -1.0R(SL) | -$2,000(SL) | Stop hit | - |
| May 13 | 14:33 UTC | USDJPY | Long | Claude Opus 4.6 | USDJPY LONG — NY AM VWAP Pullback Continuation | B | -0.25R(SL) | -$500(SL) | Stop hit | - |
| May 13 | 14:43 UTC | NAS100 | Short | Claude Opus 4.6 | VWAP/Fib Rejection Short | 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 of the week was correlation that our sizing rules did not price.
Tuesday's US30 stop was a standalone event: one instrument, one read, one structural failure. Nothing about it predicted Wednesday. Wednesday is where the pattern lives. The USDJPY long and the NAS100 short triggered ten minutes apart, and while they sat on opposite instruments, they were not opposite bets. Both leaned, directly or indirectly, on a risk-on continuation that the tape declined to deliver inside that hour. When the continuation failed to print, both positions resolved against us in quick succession. The drawdown curve does not show three separate dents. It shows one Tuesday dent and one deeper Wednesday cliff where two stops compounded.
The lesson is not that the reads were wrong. The USDJPY setup was a B. The lesson is that our risk layer evaluates each entry on its own structural merits and does not currently apply a portfolio-correlation discount when two positions would resolve on the same macro question within the same hour. That is a tuning target, named honestly below, not a defect we are hiding behind prose.
Decision highlights
The Risk Agent allowed two correlated entries, the USDJPY long and the NAS100 short, to size in within ten minutes of each other on Wednesday. Each cleared its own structural bar, but neither evaluation accounted for the fact that both positions would resolve on the same risk-on question inside the same hour. They stopped together, and the joint loss is the deepest point on the week's drawdown curve.
The Macro Agent did not veto Tuesday's US30 short, and that was the correct call given the regime read at the time. The trade still lost. This highlight is here to make a point we hold to: a non-veto that precedes a loss is not a Macro error if the regime genuinely did not warrant a veto. We grade decisions on the information available at decision time, not on the outcome.
The system declined nothing it should have taken and forced nothing it should have skipped. The honest decision review for this week is that the process ran as designed and the market disagreed three times. That is the least dramatic decision highlight we can write, and on a loss week it is also the most important one.
Claude vs GPT: who led the week.
SkyAnalyst runs multiple foundation models in parallel across its four-agent system. When two models trade the same instrument in the same week, the results are directly comparable. This is that comparison.
- NAS100-1.0R · 1 trade
- US30-1.0R · 1 trade
- USDJPY-0.3R · 1 trade
Same signals, same risk framework, different foundation model.
Six instruments, six stories.
EURUSD: no losses this window. The pair sat outside our setup criteria and produced nothing for the loss-attribution sample.
All EURUSD this week →GBPUSD: no losses this window. Cable never cleared confluence; no position was sized.
All GBPUSD this week →US30: one loss, minus 1R. Tuesday's sell-the-pullback short into the opening-range high stopped when the level it leaned on did not hold. A C-plus setup that failed the way C-plus setups fail.
All US30 this week →NAS100: one loss, minus 1R. Wednesday's VWAP and Fibonacci rejection short was the loss of the week. It stopped inside the same hour the yen long did, the deeper half of the Wednesday cliff on the drawdown curve.
All NAS100 this week →USDJPY: one loss, minus 0.25R. The week's cleanest read, a B-graded New York morning VWAP pullback continuation long, stopped shallow when the risk-on continuation it needed never printed.
All USDJPY this week →US500: no losses this window. No qualifying setup; the index produced nothing for this report.
All US500 this week →Loss of the week: NAS100 Short · -1R
Losses worth learning from
USDJPY long, May 13, B grade, minus 0.25R
What was right: The read was the cleanest of the week. A New York morning VWAP pullback continuation long on USDJPY, graded B, entered on structure the Trend Agent scored as a genuine continuation rather than a reach. The Macro Agent did not veto. Sizing was the standard 2 percent risk. Nothing about the entry was forced.
What was wrong: The continuation the trade needed never printed inside the hour. Worse, this long and the NAS100 short were the same bet wearing two instruments: both wanted risk-on follow-through, and when it failed, both resolved against us. The stop was shallow at minus 0.25R, which is the trade behaving correctly under a fast invalidation.
What we'd do same: Take the setup. A B-graded continuation that invalidates fast for a quarter-R is the system working, not failing. The change we want is upstream, in the correlation check, not in this entry.
NAS100 short, May 13, C-plus grade, minus 1R
What was right: The VWAP and Fibonacci rejection short was a coherent C-plus read. The structure was there, the Trend Agent scored it honestly as acceptable rather than premium, and the position sized to the same fixed 2 percent the rules apply to every trade regardless of conviction.
What was wrong: This was the loss of the week, a full minus 1R, and it landed ten minutes after the yen long on the same failed risk-on premise. A C-plus setup carries less margin for error than a B, and this one had none: the rejection it needed did not hold and price ran the stop. Stacked with the correlated yen entry, it turned a manageable Wednesday into the drawdown curve's low.
What we'd do same: Take a clean C-plus when the confluence clears the threshold; the system is built to trade the median setup, not only the outliers. What we would not do again is let it size without a portfolio-correlation discount against the open yen position.
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 | -2.25R | −$4,500 |
All six agents combined.
From the desk
A note on the week, before we move on.
We almost did not write this one as a standalone report, and then we remembered that the weeks we most want to skip are the ones the gate exists to catch. Three losses, 2.25R, no green prints in the attribution sample. The gate fired on its own arithmetic, lossCount and total R lost, with no override from us. That matters: we did not force a drawdown story onto a quiet week, and we did not bury a real one. The cadence is supposed to have organic holes and organic red weeks, and this was a red one.
Put the dollars on the table. A 100,000 dollar simulated account sized at 2 percent per trade sits near 140,857 dollars on a static basis through May 18, 2026, against roughly 147,309 dollars if you compound the same R through 2 percent risk. This week's give-back is approximately 4,500 dollars of that figure. The static and compounded paths diverge by about 6,500 dollars now, and that gap is not noise; it is the visible evidence that disciplined fixed-fractional sizing compounds even across the weeks it loses. A red week does not break the path. It is priced into the path.
The honest close is the unromantic one. The system did this week what it does every week: it read the tape, it sized to its rules, and three times the market declined to confirm. We found one real thing to fix, the correlation gap, and we are scoping it narrowly rather than rewriting the engine because a normal week ran red. The next report will look different. They always do.
From the desk, the SkyAnalyst Team.
What we're tuning
The one concrete tuning target this week is the portfolio-correlation gap the Wednesday stops exposed. Our Risk Agent evaluates each entry on its own structural merits and does not currently apply a correlation discount when two open or pending positions would resolve on the same macro question inside a short window. The USDJPY long and the NAS100 short were not independent draws; they were one risk-on bet on two instruments, and they stopped together. The change we are scoping is a correlation-aware sizing check that flags clustered exposure before the second entry sizes in, rather than after both have stopped.
We are explicitly not tuning the reads themselves. The USDJPY setup was a B and the system should take it again; the C-plus prints cleared their threshold honestly. Overfitting the entry logic to three stops in one week is exactly the mistake the statistics section warns against. The fix belongs in the correlation layer, scoped narrowly, and validated against the longer record before it ships.
What the numbers actually mean
Here is the framework for reading this week's numbers without flinching or spinning. The loss-attribution window carries a 34.1 percent win rate over a 91-trade sample, an average winner targeting roughly 1.17R, a current peak-to-trough drawdown of 2.4 percent, and a longest losing streak of 2 inside the window. A three-trade week that goes 0-for-3 looks alarming in isolation and is entirely unremarkable in context, and the math is worth walking through because it is the same math every legitimate fund publishes when it reports a drawdown.
Start with streaks. Van Tharp's work on R-multiples in Trade Your Way to Financial Freedom makes the core point plainly: a system with a sub-50 percent win rate will, with certainty over a long enough record, produce losing streaks that feel pathological in the moment and are statistically routine. A strategy that wins around a third of its trades expects clusters of consecutive losses as a structural feature, not a malfunction. A two-trade streak inside a single week is nowhere near the tail of that distribution. Jack Schwager's Market Wizards interviews return to the same theme from the practitioner side: the traders who survive are the ones who sized so that an ordinary streak could not end them, and who did not change the system because a normal bad run felt abnormal.
Now the arithmetic of this specific week. Three stops for a combined minus 2.25R is, by design, survivable: the risk layer holds each trade near a fixed fractional risk, so a full week of losers caps the account-level bleed at a known, small figure rather than an open-ended one. The drawdown curve confirms it. The trough sat at minus 2.4 percent and the equity walk had recovered above it by Friday. A 2.4 percent dip is not a system in distress; it is a system inside the normal operating envelope its win rate and R-target predict. We deliberately do not publish a computed Kelly fraction or an expected-drawdown number the system does not actually trade on, because a precise figure the engine never uses would be theater. The honest statement is the bounded one: a 34 percent win rate carries routine losing clusters, fixed fractional sizing makes those clusters non-lethal, and one week of N equals three is too small to update a 91-trade base rate. That frame does not change whether the week was red or green.
At a Glance
Drawdown questions
Three losses and zero wins this week. Is the system broken?
No. The loss-attribution sample for this report is the three stops; the equity walk that the drawdown curve uses shows Friday recovered the ground and printed a fresh high. A 0-for-3 stretch on a system that wins roughly a third of its trades is well inside the normal distribution of streaks, not evidence of breakage.
Why publish a losing week at all?
Because every legitimate fund publishes its drawdowns, and a journal that only shows green weeks is marketing, not a record. The system carries plus 20.43R for the year through May 18, 2026; this week gave back 2.25R. Showing both is the only honest way to present the ratio.
What is the single thing you are changing after this week?
One thing: a portfolio-correlation check in the Risk Agent so two positions that would resolve on the same macro question inside a short window do not both size in at full risk without a clustered-exposure discount. We are not changing the entry reads, which cleared their thresholds correctly.
How bad was the drawdown in account terms?
The peak-to-trough drawdown bottomed at 2.4 percent and had recovered by Friday. On a 100,000 dollar simulated account at 2 percent risk, the week cost roughly 4,500 dollars, a bounded figure by design because the risk layer holds each trade near a fixed fractional risk.
Were any of these losses caused by a rule violation?
No. Every stop this week was a setup the rules cleared and the tape declined to confirm. None were forced entries or skipped vetoes. The distinction between a rule that produced a loss and a rule that was broken is the whole point of the report, and this week was entirely the former.
Trade with the system that publishes its drawdowns.
Subscribers receive every signal — winners and losers — 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: +20.43R YTD across 99 trades, see stats strip.
From the SkyAnalyst Journal
EURUSD Short on May 15: Six Waits, One Enter, a Clean TP1 Fill
A risk-off Euro short where the system scored six consecutive waits in the low 40s, then flipped to enter at 62 percent, and the position closed TP1 for +2.00R (TP1) with zero recorded drawdown.
May 11-17, 2026: Seven-Three Week, +5.96R Net Across a Model Upgrade
Ten canonical trades, seven winners, three losers, +5.96R net at the TP1 baseline. Tuesday and Wednesday ran on Claude Opus 4.6, Friday switched to Opus 4.7, and GBPUSD came online as a new instrument and won both its trades.
GBPUSD Short on May 15: Five Waits, One Enter, a Clean TP1 Fill
A risk-off Cable short where the system scored five consecutive waits in the low 40s, then flipped to enter at 72 percent, and the position closed TP1 for +1.89R (TP1) with zero recorded drawdown.