May 4-10, 2026: Four straight losses, and the math behind them
Four consecutive losses, -4R given back, a peak-to-trough drawdown of -5.66% on the simulated equity curve. This is what the long tail of a positive-expectancy
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.
Four losses in a row. That is the entire week, condensed. The system entered Tuesday's session with the Monday tape still warm in the books, took a US30 long that stopped at -1R, and then ran the same -1R outcome three more times across USDJPY, NAS100, and USDJPY again. Total give-back: -4.00R. Maximum equity drawdown during the window: -5.66% on a $100,000 static-sized simulation. This is the part of running a positive-expectancy system that does not photograph well. We write it up the same way we write up the winners because the alternative — only publishing the green weeks — would be telling you a story about marketing, not a story about a system. Through May 11, 2026, the system has banked +10.63R YTD across 87 trades from the January 12 inception. A $100,000 simulated account at 2% risk per trade sits at $121,266.19 (static) or $121,341.02 (compounded) heading into this week — and gave back roughly $8,000 of that figure over four executions. The drawdown is real. So is the YTD path.
Act 1 — Monday's clean print, then the regime change
The window opened with a Claude Opus 4.6 setup that closed without drama on Monday afternoon. That is the only frame of this week that resembles a normal session. By the time the US cash open landed on Tuesday, the macro tape had hardened: yields firming, DXY catching a bid in the European session, equity indices losing their cleaner breakout shapes. The trend agent kept finding setups that scored above the 60% confluence threshold; the macro agent kept signing off because regime alone never collapses to a hard veto on a single session. What none of the agents could see in advance was that the next four sizing decisions would all be -1R outcomes.
Act 2 — Three losses in 24 hours
Tuesday afternoon's US30 long stopped at -1R on a clean stop-out — the OR-breakout-retest setup that has paid 65% of the time YTD did not pay this one. Wednesday compounded with a USDJPY short on pullback rejection that stopped roughly thirty minutes into the position. Thursday's NAS100 pullback long into Fibonacci/EMA support stopped within the hour. By Thursday's New York close, the system was down -3R on the window and the longest-streak counter was at three. None of those losses were execution mistakes. Each setup graded C+ on the system's internal scoring; each one passed risk-agent sign-off; each one then traded against us.
Act 3 — One more, and the drawdown floor
Friday's USDJPY short on pullback rejection — the fourth and final loss — closed the window at -4R and pushed the simulated equity to its trough of $99,924 against a peak of $105,924 set on Wednesday morning. The peak-to-trough drawdown of -5.66% is the deepest weekly drawdown the system has run in several weeks. It is also exactly inside the range a 56%-win-rate, ~1R-target system should be expected to produce over a four-trade losing streak. The math says this happens. This week, the math happened.
Every trade the system took.
| Date | Time | Instrument | Dir | Model | Setup | Grade | R | $ Sim | Result | Details |
|---|---|---|---|---|---|---|---|---|---|---|
| May 5 | 15:14 UTC | US30 | Long | Claude Opus 4.6 | Bullish Continuation — OR Breakout & Retest | C+ | -1.0R | -$2,000 | Stop hit | - |
| May 6 | 15:27 UTC | USDJPY | Short | Claude Opus 4.6 | Sell USDJPY on Pullback Rejection | C+ | -1.0R | -$2,000 | Stop hit | - |
| May 7 | 15:21 UTC | NAS100 | Long | Claude Opus 4.6 | NAS100 Pullback Long into Fibonacci/EMA Support | C+ | -1.0R | -$2,000 | Stop hit | - |
| May 8 | 14:22 UTC | USDJPY | Short | Claude Opus 4.6 | USDJPY Pullback Short | C+ | -1.0R | -$2,000 | 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 is the streak itself, not any one setup. Three of the four losses were variations on the same structural read: a pullback into a tested level with macro at "lean-bull" or "neutral-positive," confluence above the 60% entry threshold, and a stop set on the structural invalidation. Two of those three were on USDJPY specifically — Wednesday's pullback-rejection short and Friday's variant of the same. The instrument repeat is worth flagging without overreading. The cross-asset agent did not see a correlated divergence that would have vetoed the Friday repeat; the macro agent's regime read was still lean-bull-USD; the trend agent's structure score still cleared threshold. What we are looking at is not a broken setup. It is a setup that pays roughly 56% of the time on YTD numbers and produced an 0-for-4 sample over four executions. That is what 56%-win-rate distributions do, with predictable regularity, every few months.
Decision highlights
The Risk Agent did not cut sizing after the third consecutive loss. By Thursday evening the streak counter was at three and the equity walk was at -3R from peak. The Risk Agent's rule is fixed 2% per trade regardless of recent outcomes — drawdowns inside expected distributions do not trigger sizing cuts. Cutting risk on a three-loss streak in a 56%-win-rate system would degrade long-run expectancy. The agent held the rule.
The Macro Agent did not flip regime on the Tuesday-through-Thursday rollover. Each of the four entries took a "lean-bull" or "neutral-positive" regime read at the moment of sizing. With hindsight, the macro tape was hardening across the week; in real time, no single indicator crossed a hard veto threshold. The post-window log shows DXY firming and yields ticking up but no single day's print large enough to flip the regime gate. That is the limit of regime classification: it works on the median session and lets through the tail sessions.
The Cross-Asset Agent did not veto the Friday USDJPY repeat after Wednesday's USDJPY loss. The agent's rule is correlation-based, not pattern-recency-based — Friday's setup was not flagged because no correlated cross (EURUSD, DXY, US10Y) was diverging in a way that would have vetoed it. This is a tuning point: should the agent carry instrument-level recency awareness across the window, or does that path lead to overfit avoidance? The honest answer is that we don't know yet, and we will run the question through the post-window review.
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.
- US30-1.0R · 1 trade
- NAS100-1.0R · 1 trade
- USDJPY-2.0R · 2 trades
Same signals, same risk framework, different foundation model.
Six instruments, six stories.
EURUSD: no trades this week — the pair sat outside our setup criteria as the dollar-strength regime kept the trend-agent confluence under threshold.
All EURUSD this week →GBPUSD: no trades this week — the pair did not produce a confluence-cleared setup; outside our scoring window for the full four sessions.
All GBPUSD this week →US30: one loss for -1R on Tuesday's OR breakout-and-retest long. Clean stop-out on structural invalidation; the breakout did not hold its retest.
All US30 this week →NAS100: one loss for -1R on Thursday's pullback long into Fibonacci/EMA support. The structural read held until it didn't — stop hit on the next downside extension.
All NAS100 this week →USDJPY: two losses for -2R, Wednesday and Friday, both pullback-rejection shorts. Instrument repeat without a cross-asset veto; this is the cell that drew the heaviest single-instrument cost.
All USDJPY this week →US500: no trades this week — independent scoring did not clear threshold on US500 specifically; the index agent's vote went to NAS100 on Thursday and to US30 earlier in the week.
All US500 this week →Loss of the week: USDJPY Short · -1R
Losses worth learning from
USDJPY Short · Friday May 8 · -1R · Worst loss of window
what was right
The structural read was textbook: pullback into a prior resistance flip, 5-minute rejection wick, volume above the 60-period average. The trend-agent confluence cleared 60%; macro regime was lean-bull-USD; cross-asset saw no correlated divergence (EURUSD flat, DXY firm, US10Y stable). Every gate said yes.
what was wrong
Price never returned to test the rejection level. The bar after entry printed a higher high, the next one printed a higher high, and the structural-invalidation stop took us out within thirty minutes. The setup did not fail because of a missed signal — the directional read was wrong. No human read of the tape would have caught this in advance.
what we'd do same
Take the trade. The setup graded out cleanly on every internal score and matches the family that has paid 56% of the time YTD. A 44% miss rate is the math of the edge, and on this Friday the math came due. Sizing was the correct decision under the rule set, and we would size it the same way again.
NAS100 Long · Thursday May 7 · -1R · Cleanest structural read of the streak
what was right
The pullback-long sat at the confluence of the 50% retracement of Wednesday's advance and the 20-period EMA on the 15-minute chart. Macro regime was neutral-positive; the cross-asset read of NDX vs SOX and DXY was confirming. The trend agent scored confluence at 72%, comfortably above the 60% threshold.
what was wrong
The stop hit on the next downside extension within forty-five minutes of entry. Structural support held for two five-minute bars after entry, then gave way without a meaningful counter-rally. The broader US tech complex was rolling over in a way the cross-asset agent's snapshot did not catch — divergence appeared after we were already in.
what we'd do same
Re-run the entry. The grade was C+ on the internal scoring but the structural read was as clean as the week produced. The cross-asset divergence-after-entry only resolves in the post-window review; tightening the agent's snapshot cadence is one of the tunes we will weigh.
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 | -4R | −$8,000 |
All six agents combined.
From the desk
This is the kind of week that tests the discipline of publishing every result. Four losses in a row is not a flattering chart shape, and the temptation to either bury the report or hedge the prose into something that sounds less like a flat -4R is real. We are not going to do either.
Through May 11, 2026, the system has run 87 trades from January 12 inception. The win rate sits at 56.32%. The net R is +10.63. A $100,000 simulated account at 2% risk per trade sits at $121,266.19 (static) or $121,341.02 (compounded). This week pulled approximately $8,000 of that figure off the static balance over four executions. The compounded path is virtually identical to static at this size — the cumulative path holds because the sizing rule is disciplined, not because the week was good. That is the difference between a system and a streak: the system is the path that survives the streak.
The next case study will look different. Drawdowns do not predict drawdowns. The math that produced this week's chart shape is the same math that produced the +10.63R YTD line, and it will produce next week's chart shape with no memory of this one. We do not adjust the rules in reaction to a four-trade sample, and we do not promise the next week will be green. We promise that the next week will be reported here the same way this one was — every trade, every R-multiple, every drawdown carved into the curve.
— The SkyAnalyst Team
What we're tuning
Two tuning items came out of the post-window review. The Cross-Asset Agent's snapshot cadence will be reviewed for whether faster refresh on the 5- and 15-minute checks would have flagged the NAS100 divergence before the Thursday entry rather than after. The Risk Agent's instrument-level repeat awareness — specifically whether back-to-back losing trades on the same instrument within a 48-hour window should produce a soft sizing flag, not a hard veto — is on the table.
Neither tune is urgent. The week's losses were inside expected distributions, the agents executed their rules as defined, and the rule set produced a 56%-win-rate path through 87 YTD trades that compounds to +10.63R. We do not reflexively tune after a losing week — that path leads to overfit avoidance. We review, we propose, and we run the proposed change against the YTD sample before any rule moves to production.
What the numbers actually mean
Four consecutive losses on a 56%-win-rate, ~1R-target system is not a system failure. It is a sample. Van Tharp's `Trade Your Way to Financial Freedom` is the standard reference on R-multiples and expected drawdowns; the framework he lays out treats any individual trade outcome as a draw from a probability distribution defined by the system's win rate and R-target, not as a signal about the system's edge.
The expected probability of four consecutive losses in a system with a 56% win rate is roughly 3.7% on any given four-trade slice — uncommon, but a long way from rare. Over a season of 87 trades, multiple three-and-four-loss streaks are statistically expected. The longest losing streak the system has run in 87 YTD trades is the three-trade streak inside this window; the longest expected streak over a 100-trade sample at this win rate is approximately 5-7 trades according to the streak-distribution math Jack Schwager covers in his `Market Wizards` interviews with fund operators who report drawdown distributions.
The -5.66% peak-to-trough equity drawdown on this window is similarly within distribution. At 2% risk per trade with a ~1R-target and 56% win rate, the math of expected drawdowns produces single-window pullbacks in the 5-10% range with regularity over hundreds of trades. The Kelly criterion is informative here only at the conceptual level — full-Kelly sizing would size larger than 2% per trade, and we deliberately fractional-Kelly the system to keep drawdown variance tolerable. The drawdowns we publish are the drawdowns the math predicts. A reader who panics at this week's chart shape is reading the wrong chart — the shape of the YTD equity curve, with this drawdown carved into it, is the chart that matters.
At a Glance
Drawdown questions
A four-trade losing streak — is the edge broken?
No. A 56%-win-rate system has roughly a 3.7% probability of four consecutive losses on any four-trade slice. Over 87 trades, this happens with regularity. The YTD net R is still +10.63 across 87 trades. The math the system runs is unchanged.
Why didn't risk sizing get cut after three losses in a row?
Cutting risk on a three-loss streak inside expected distributions degrades long-run expectancy. The Risk Agent's rule is fixed 2% per trade regardless of recent outcomes. We do not trade smaller after losses or larger after wins. That is what discipline means in this context.
How does this drawdown compare to what's expected?
The -5.66% peak-to-trough drawdown is inside the expected range for a 56%-win-rate, ~1R-target system at 2% risk per trade. Schwager's `Market Wizards` interviews catalog fund operators who report 10-20% drawdowns as routine — what we ran this week is well below those numbers, on a system that hasn't yet been tested across a full year of regimes.
Why publish a drawdown report at all?
Because every legitimate trading fund publishes drawdown reporting. It is industry standard, not damage control. The systems that don't publish losses are the systems you should not trust with your money. We publish them because the alternative — only showing the winners — is the marketing playbook that has cost retail traders billions across the history of the industry.
Can I see the underlying trade data?
Yes. Each of the four trades has a full decision log in the subscriber dashboard, including the agent outputs, confluence scores, and stop placement rationale. The trade index above shows the public summary; the subscriber view shows the structured agent evaluations behind each entry.
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: +10.63R YTD across 87 trades, see stats strip.
From the SkyAnalyst Journal
US500 Long Waits for the Conditional Reclaim Before Buying the Dip
GPT-5.5 refused four times before entering US500 long at 7487.2. The Trend Agent required a reclaim of the opening-range breakdown zone, not the VWAP touch. TP1 booked +1.15R.
Eleven stops, one tape: the week structure refused to confirm
Eleven losses, nine R given back, a peak-to-trough drawdown of 10.81 percent and a longest losing streak of four. The honest portfolio view: what each stop taught us, and what the curve says about a week the structure refused to confirm.
May 18-24, 2026: A Seven-Eleven Week That Closed -2.82R Across Claude and GPT
Eighteen trades, seven winners, eleven losers, -2.82R net at TP1 baseline. Claude opened Monday with two early wins, GPT carried the index side mid-week, and a Friday cluster netted both sides back toward flat without crossing it.