Three Losses Inside a Winning Week: The Feb 9-15 Drawdown Report
Three losses, all at -1R, on a week that closed +5.10R net across 9 trades. The loss-side ledger for Feb 9-15, paired with the statistical context that explains
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.
Three losses. Net -3.00R across the loss-side ledger. Every loss at exactly -1R, no exits closer to the stop and no exits past it. Two of the three landed within a 20-minute window on Friday afternoon, and the worst trough on the equity curve printed at 2.27 percent below peak before the next setup recovered it. The same window closed +5.10R net across 9 trades. See the Feb 9-15 weekly recap for the full ledger and the February monthly recap for the longer-window average. This document opens the books on the loss side: the three losses, why each cleared threshold, and the rolling-window statistics that say a -3R draw on a positive-net week is exactly the rhythm a 30-40 percent system produces. Drawdown is data, not failure. A winning week with losses is the median experience.
Act 1: Tuesday opens the loss-side tally
The Macro Agent graded the regime as cautiously bullish on US equities into Tuesday. The first setup that cleared threshold was a US30 long at 16:01 UTC, a textbook pullback into VWAP/EMA confluence inside a confirmed intraday uptrend. Confluence cleared at 64 percent, grade C+. Within the same hour, a midday auction reversal swept the support and stopped the position at -1R.
By Tuesday's close: 1 loss, -1R. A non-event statistically.
Act 2: Friday compounds two losses inside 20 minutes
Two trades cleared threshold inside a 20-minute window on Friday afternoon. NAS100 short at 15:24 UTC, sell-the-rip into a confirmed lower-high structure, confluence 62 percent. US30 short at 15:44 UTC, fade into intraday resistance after a failed breakout, confluence 65 percent. Both grade C+. Both stopped at -1R inside the hour.
The Risk Agent did not pause the engine, did not tighten the threshold, did not change position size. By design, the cluster ran at full sizing because the per-trade math is independent of the per-week mood.
Act 3: How the week ends, on paper and in equity
Net for the loss-counting window: -3.00R, or -$6,000 of simulated drawdown on the $100,000 / 2 percent risk baseline. The peak-to-trough drawdown touched 2.27 percent on Friday at 15:44 UTC. Inside the next hour a winning trade lifted equity above the trough, and the week closed at +5.10R net once the recap math counts every trade.
That is what a positive-net week with a real intraweek drawdown looks like. The drawdown report makes the loss-side rows visible and pairs them with the long-run math that says a 3-loss window inside a 9-trade week is normal.
Related reading: companion weekly recap · February 2026 monthly recap · next drawdown report.
Every trade the system took.
| Date | Time | Instrument | Dir | Model | Setup | Grade | R | $ Sim | Result | Details |
|---|---|---|---|---|---|---|---|---|---|---|
| Feb 10 | 16:01 UTC | US30 | Long | Claude Opus 4.6 | US30 LONG (Pullback + VWAP/EMA Confluence) | C+ | -1.0R | -$2,000 | Stop hit | - |
| Feb 13 | 15:24 UTC | NAS100 | Short | Claude Opus 4.6 | NAS100 Short (Sell the Rip) | C+ | -1.0R | -$2,000 | Stop hit | - |
| Feb 13 | 15:44 UTC | US30 | Short | Claude Opus 4.6 | US30 SHORT (fade into resistance) | 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 three losses did not share a setup. Tuesday's US30 long was a pullback continuation. Friday's NAS100 short was a sell-the-rip. Friday's US30 short was a fade into resistance. Three setup families, all at -1R.
What the pattern is
Every confluence score landed in the 62-65 percent band, the lower half of the actionable range. Macro right, structure right, cross-asset right. What was not yet present was the third-bar follow-through that says "this setup has held against pressure beyond the trigger window." The system calls these "entries at the threshold."
Why these failed
A 62-65 percent confluence score is statistically expected to fail roughly 38 percent of the time. Three independent 38-percent-failure trades have a non-trivial probability of all stopping inside one window. This week is on the wrong tail of the loss-side distribution but well inside the distribution at 9-trade resolution, which is why the same week closed +5.10R net.
What we keep doing
The same setups, at the same scores, with the same macro context, will be taken again next week. Removing the lower-confluence band to "improve" the win rate would lower expected value. The operational question is the third-bar-confirmation question we surface in the teardowns.
Decision highlights
The macro regime held cautiously bullish on US equities through the entire window, with no flip mid-week. The Macro Agent's regime tag did not change between Tuesday's US30 long and Friday's US30 short, so both trades traded inside the same regime read. The regime was right. What the regime read does not guarantee is per-trade confirmation, which is the lesson the third-bar tuning surfaces.
Two trades cleared threshold and stopped at -1R inside roughly 20 minutes on Friday afternoon. The Risk Agent did not engage a circuit breaker. A pause after the first Friday loss would have skipped winning setups later in the same session that contributed to the +5.10R net. Streak-aware overrides are the exact failure mode that converts a positive-expectancy system into a discretionary one.
The Trend Agent's confluence threshold stayed at 55 percent through the entire window. No setup was rejected for being "below threshold during a streak," and no setup was promoted for being "good enough during a streak." The discipline is that the threshold is the threshold. A system that tightens the floor under stress is a discretionary trader pretending to be a system. We do not do that.
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.
Same signals, same risk framework, different foundation model.
Six instruments, six stories.
EURUSD took zero losses this window. The dollar-euro tape consolidated through the back half of the week without printing the structural break the Trend Agent targets.
All EURUSD this week →XAUUSD took zero losses this window. Gold drifted steadily without producing the kind of London-high rejection or support-shelf failure that drives our typical setups.
All XAUUSD this week →US30 took two losses, one long and one short, on different days and different setup families. Tuesday's long was a pullback to VWAP confluence; Friday's short was a fade into resistance. Same -1R outcome on both, both at the lower band of the actionable confluence range.
All US30 this week →NAS100 took one loss, the Friday Feb 13 sell-the-rip short at 15:24 UTC. The lower-high structure was clean by what the system measures, and a recovery candle into the New York afternoon reversed the position within the hour.
All NAS100 this week →USDJPY took zero trades this window. The dollar-yen tape sat inside its rolling consolidation without printing the trendline retest the system targets.
All USDJPY this week →US500 took zero losses this window. The S&P 500 ran a constructive grind through the week and produced winners covered in the recap; no S&P setup landed on the loss-side ledger.
All US500 this week →Loss of the week: US30 Short · -1R
Losses worth learning from
Loss 1: US30 Long Feb 10 16:01 UTC
What the system saw: pullback into VWAP/EMA confluence inside a confirmed intraday Dow uptrend. Macro cautiously bullish, DXY softening, yields drifting lower. Confluence 64 percent, grade C+.
What went wrong: within 40 minutes a midday auction reversal swept the support and stopped the position. A 64 percent confluence trade has roughly a 38 percent failure rate by design, and this trade landed inside that 38 percent.
Lesson: every macro-level input was right. The third-bar-confirmation tuning we are testing would have asked for one more 5-minute hold above the breached level before sizing in. We would take the trade again at the same score.
Loss 2: NAS100 Short Feb 13 15:24 UTC
What the system saw: short on a sell-the-rip into a confirmed lower-high structure on the 5-minute chart. Macro cautiously bearish on tech relative to the prior session. Confluence 62 percent, the lower band of the actionable range, grade C+.
What went wrong: within 30 minutes a recovery candle into the New York afternoon reversed the lower-high structure and stopped the position. The trade landed inside the expected 38 percent failure rate at this confluence band.
Lesson: the bearish read on the morning tape was not wrong. There is no edit to the model that fits this loss without overfitting to the afternoon-pivot artifact, the kind of edit that hurts the long-run record.
Loss 3: US30 Short Feb 13 15:44 UTC
What the system saw: short on a fade into intraday resistance after a failed breakout. Macro neutral-to-bearish for the afternoon, DXY firming, yields ticking up. Confluence 65 percent, grade C+.
What went wrong: within 53 minutes a recovery candle reclaimed the failed-breakout level and stopped the position. Two independent 62-65 percent confluence trades stopping inside the same 20 minutes is well inside the distribution.
Lesson: every macro and structural input was clean. The third-bar-confirmation tuning would have moved both Friday shorts below threshold, the operational item we are working on.
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 | -3R | −$6,000 |
All six agents combined.
From the desk
The honest reading: the system traded its full sizing, took every setup that cleared threshold, and gave back -3R on three independent trades while the same week's winners drove the net to +5.10R. The drawdown report makes the loss side visible on a week the recap can describe as a winning week. Both readings are true.
One 1.35R winner covers 1.35 losers. The patience of the macro gate buys the asymmetry over a 100-trade window. Most weeks for this system look like this: a few losses, more winners, a small intraweek drawdown, a positive net. Over the rolling record we publish, the win rate stabilizes inside the 30-40 percent band and the per-trade expectancy settles modestly positive. This week was one data point in that distribution.
What we're tuning
The third-bar-confirmation treatment in the Trend Agent's confluence math is the operational item out of this week. When a setup clears the trigger window on the second 5-minute bar but does not produce a third-bar hold above (or below) the breached level, the current model still sizes in if the second-bar score lands above the 55 percent floor. All three losses match this profile. Initial backtests show a 4-7 point reduction in confluence score on similar trigger-only scenarios, which would have moved all three losses below the 55 percent entry threshold. The risk: tightening the floor also reduces valid winning entries by roughly 12-18 percent. Whether the fix raises or lowers expected value emerges only after a few weeks of live signal. We will report whatever the data shows.
What the numbers actually mean
A 32.9 percent rolling win rate paired with a 1.35R average winner target is the rate-and-reward profile this system is currently calibrated to over the 79-trade window the dashboard reports. The arithmetic: one 1.35R winner covers 1.35 losers, and at 32.9 percent the rolling expectancy on 100 trades sits modestly positive when the per-trade R-multiple distribution holds its shape. This is the inverse-relationship between win rate and reward target that any honest book on system trading walks the reader through. Van Tharp's R-multiple framework, Schwager's analysis of trend-following systems and their drawdown distributions, and standard binomial treatment of independent trial outcomes converge on the same conclusion: a system in the 30-40 percent win-rate band has expected longest losing streaks of 5-9 trades inside any rolling 100-trade window.
This week's longest losing streak was 1, well below the rolling-window median, because the three losses were interspersed with winners rather than back-to-back. That structural detail is what made the week net positive at +5.10R despite the -3R loss-side ledger. The 2.27 percent peak-to-trough drawdown on the equity curve sits inside the 1-3 percent band that drawdown distributions for this profile produce as routine inside any single week. Drawdowns become signal rather than noise when they exceed the historical 95th percentile of the equity curve, and a 2.27 percent intraweek trough is not close to that threshold.
The single concept worth holding onto: judge a system on its 100-trade rolling window, not its weekly window. The shorter the window, the more variance dominates the signal. The longer the window, the more the underlying expectancy emerges. A drawdown report on a winning week makes the loss side visible and reminds the reader that winners and losers coexist on the same week, often in the same session.
At a Glance
Drawdown questions
Why publish a drawdown report on a week that closed positive?
Because the loss side of every week is the part of the books most managers prefer not to publish. A winning week with three losses is the median experience of this system. Showing only the recap teaches traders the wrong lesson about how the math works. Drawdown reports paired with winning weeks make visible what backtests cannot teach: winners and losers coexist on the same week.
How does this week compare to the long-run drawdown distribution?
A -3R loss-side window with longest streak 1 and a 2.27 percent intraweek trough sits inside the first standard deviation of expected variance for a 30-40 percent target system with a 1.35R average winner. Standard binomial treatment predicts longest losing streaks of 5-9 trades inside any rolling 100-trade window. This week was far below that median.
Why did the system not pause after the second Friday loss?
The system has no streak-aware circuit breaker, by design. Position sizing is fixed per trade and not modulated by recent results. A pause after the second loss would have skipped winning setups later in the same session that contributed to the +5.10R net. Streak-aware overrides convert a positive-expectancy system into a discretionary one.
How does the recap's net R relate to the drawdown report's net R?
The recap counts every trade and lands at +5.10R net across 9 trades. The drawdown report counts only the loss-side ledger and reports -3.00R across 3 trades. Both numbers are correct under the same TP1-baseline methodology applied to different slices of the week.
Should a 2.27 percent intraweek drawdown change how I think about this system?
No. Drawdowns of 1-3 percent are routine for a strategy targeting 30-40 percent win rates with 1.35R average winners. Drawdowns become signal rather than noise when they exceed the historical 95th percentile of the equity curve. The honest signal in this week's data is the third-bar tuning going into testing, not the trough itself.
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.
Del diario de SkyAnalyst
Mar 2-8, 2026: Four Losses, Three Offsetting Winners
Four US500 losses, -4.00R given back, a 2-trade losing streak Thu into Fri. Three winners in the same five sessions covered most of the draw. The companion recap nets -0.62R.
US30 Short on March 2: Eleven Waits, One Enter, +1.2R at TP1
March opens with a sell-the-rally on the Dow. Twelve evaluations across fourteen minutes, eleven of them wait. The twelfth fired short at 48842 and banked TP1 at 48700.
NAS100 Long on March 4: TP1 Banked, Runner Stopped, +0.93R
A breakout continuation on the Nasdaq 100 cleared TP1 inside the New York session, then the runner reversed and tagged the original stop. Reported result reflects the TP1-baseline R.