USDJPY Long Books a Counter-Macro TP1 on Tight VWAP-Fib-EMA Support
SkyAnalyst AI journal entry: USDJPY Long on May 13, 2026 closed +0.87R on TP1. Full workspace view, decision log, and AI reasoning, unedited.
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. That’s what makes the system auditable — and it’s what this case study will show, step by step, on a specific setup the trend agent almost passed on.
The macro setup the morning gave us
May 13 was a dollar-strong session across the bloc. The other GPT-5.5 trades that day (EURUSD short, US30 short, GBPUSD short) all ran with the macro. The USDJPY long was the exception.
Why the macro lean was bearish for dollar bloc
The ten-year yield was elevated and the DXY was firm, which is the dollar-strong setup. Under normal interpretation, dollar-strong against the bloc translates to USDJPY long (since USD is the buy leg). But the Macro Agent's lean_bear tag was on the FORMER dollar bloc, not on the JPY pair specifically. JPY weakness has its own carry-driven dynamic that can run independently of pure dollar strength. The cross-asset alignment was therefore ambiguous on USDJPY rather than directly negative.
The local structure was the trigger
Three independent support references coincided at 157.817: the session VWAP, the 38.2 percent Fibonacci retracement of the morning's range, and the 5m EMA cluster. When price tested that level on the morning's first pullback and held, the structural read was clean even though the macro lean was unsupportive. The Trend Agent's gate allows counter-macro entries when the local structure clearance compensates for the macro hedge. See our May 15 USDJPY long pullback for the macro-supported version of the same setup shape.
NEUTRAL at 58 percent is the gate floor
The Trend Agent's confidence floor for any entry is 55 percent. NEUTRAL at 58 percent is just above that floor, which means the Risk Agent applied conservative sizing. The TRANSITIONING regime tag on top dropped position size further to 0.75 percent equity. The trade ran with the lowest sizing the system will allocate to a winning setup.
The pattern we were trading
This was a textbook example of what professional traders: a VWAP / EMA / Fib support cluster long. Three independent support references combine on a single level, the first test holds, and the candle close back above confirms the entry.
Why three references at the same level matter
A single support reference (a VWAP, an EMA, a Fibonacci level on its own) is a weak signal. Markets routinely test and break individual support references intraday. A cluster of three independent references at the same level is structurally different: it forces three communities of traders (VWAP mean-reversion algos, EMA crossover systems, Fibonacci-watching discretionary traders) to all read the same line as support simultaneously. A failure to break that combined cluster on the first test is meaningful.
Counter-macro setup mechanics
The system allows counter-macro setups when local structure is clean enough to offset the macro hedge. The Trend Agent's confidence is capped lower (here 58 percent versus 80-plus on a macro-aligned setup), the regime defaults to TRANSITIONING, and the Risk Agent reduces position size. The setup is taken because counter-macro setups, when structurally clean, have positive expectancy at the reduced size. Refusing them entirely costs us the TP1-clean wins this kind of trade tends to produce.
Why counter-macro trades rarely extend
Counter-macro setups produce TP1 hits but rarely reach TP2 or TP3 in our internal data. The reason is mechanical: the local support holds for the first bounce, but the macro tape continues to pull the instrument in the opposite direction. The bounce exhausts at TP1 and momentum reverts to the macro lean. The May 13 USDJPY long is the canonical version of this shape: clean TP1, no extension.
Why we still take them
A 0.87R hit at 0.75 percent equity is $1,305 on the simulated $100,000 account. It is not a hero number. It is a clean number. The system's edge comes from many clean numbers compounding, not from chasing TP3 hits. Counter-macro TP1s are a meaningful contribution to the per-month R total because they happen with structural reliability when three-reference clusters present.
The 13-pip stop
The stop sat at 157.742, seven and a half pips below entry. That is tight by USDJPY standards. The Risk Agent allowed it because the entry sat at the support cluster: any meaningful break below 157.742 would invalidate all three references simultaneously, and the long thesis would be cleanly wrong. We do not size stops to "give the trade room"; we size them to where the chart says the read is invalidated.
Pattern catalogue
We trade nine setups across forex and indices: NY AM continuation, NY AM session pullback, London continuation, opening-drive rejection, VWAP reclaim, range-extreme fade, breakout-retest, Asian range break, and structural failure. Each is gated by its own confluence rules. The Trend Agent does not pick a favorite. Different days produce different setups, and the system is dynamic, not dogmatic. It doesn't favor any single strategy.
Decision log
At 16:02 UTC on May 13, the GPT-5.5 Trend Agent issued the entry decision on a single trigger event. The recorded decision log shows ENTER without intermediate WAIT evaluations: the chart trigger (a 5m candle close back above the 157.817 VWAP/Fib/EMA support cluster after a successful retest) fired on the first eligible bar. The Macro Agent flagged the counter-macro nature of the setup with a lean_bear tag at 61 percent, and the Trend Agent applied its counter-macro confidence cap, landing at NEUTRAL 58 percent. The Cross-Asset Agent confirmed JPY weakness was not severe enough to override the local structure. The Risk Agent computed entry 157.817 (post-retest reclaim), stop 157.742 (below the support cluster plus 7.5 pip buffer), TP1 157.882 (0.87R prior intraday high extension). TP2 and TP3 were not tracked given the counter-macro classification. Final confidence: 58 percent. Decision: ENTER long at 0.75 percent equity under TRANSITIONING.
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 |
|---|---|---|
| Stop hit (invalidated) | -1R | −$2,000 |
| TP1 hitActual | +0.87R | +$1,740 |
| TP2 hit — not tracked | +0R | +$0 |
| TP3 hit (max potential) — not tracked | +0R | +$0 |
All six agents combined.
What this trade taught us about counter-macro setups
The system can take a long when the macro is lean_bear and the trade still wins, as long as the local structure is unusually clean. Three independent support references at one level is unusually clean. The May 13 USDJPY long is the canonical example.
Counter-macro setups are not opportunistic
We do not take counter-macro trades because we disagree with the macro tape. We take them because the local structure produces a clearance signal that the gate accepts. The Trend Agent's confidence is capped, the regime defaults to TRANSITIONING, and position size is reduced. The system is not betting against the macro; it is harvesting a structurally-clean local bounce inside a broader trend it expects to resume.
The realized R is the full story on TP1-only trades
The realized +0.87R (TP1) and the full-potential +0.87R (TP1) are the same number on this trade. There is no gap. The reader sees exactly what the broker booked, and the published track record reflects the actual ledger entry. This is the most common shape of winner in our system, and it is the cleanest possible reconciliation between the hero figure and the running tally.
Why we report counter-macro trades the same as macro-aligned
There is no editorial markdown for counter-macro entries. The case study reports the same data points (entry, stop, TPs, confidence, regime) and the same realized-versus-full-potential disclosure. The reader sees that NEUTRAL confidence at 58 percent under TRANSITIONING is the lowest-quality grade the system trades, and the journal does not pretend the setup was something else. See our same-day GBPUSD short for a macro-aligned TRANSITIONING setup with a meaningfully higher confidence read.
What we changed in our notes after this trade
Two adjustments from the May 13 USDJPY post-trade review.
Counter-macro entry tag
We are adding a "counter-macro" metadata tag for trades where the Trend Agent's directional read is opposite the Macro Agent's lean tag. The hypothesis is that counter-macro trades hit TP1 at rates similar to macro-aligned trades, but rarely extend to TP2. Six months of post-tagged data will tell us if the per-trade R distribution differs meaningfully.
Three-reference cluster as a setup type
The May 13 USDJPY entry used VWAP, Fibonacci, and EMA as confluence references on a single level. We are starting to track how often three independent references coincide within ten pips on a single setup. If three-reference clusters predict TP1 hit rate better than two-reference clusters, that is an input the Trend Agent should weight differently.
One number you do not see in the journal
The trade tested back to roughly 157.788 about ninety minutes after entry, which translates to a brief -0.40R unrealized drawdown before the move resumed. That figure does not appear in the journal because the broker never closed the position there. We mention it here because the retest is meaningful: a counter-macro setup that does not test its support cluster at least once before extending is more likely to fail than one that does.
At a Glance
What this teaches about AI-driven trading
How does the system decide to take a counter-macro setup?
The Trend Agent gates entry on a confluence score. When local structure produces a clean cluster (three or more independent references coinciding within a tight band), the gate can clear even when the Macro Agent's lean is opposite. The Trend Agent's confidence is capped at NEUTRAL or low-BEARISH/low-BULLISH in those cases, and the Risk Agent reduces position size. We do not take counter-macro entries on weak local structure.
Why are counter-macro trades rare in our journal?
Most setups in our nine-pattern catalogue require macro alignment as a baseline gate. The Macro Agent's lean tag has more weight than the Trend Agent's local read for most patterns. Counter-macro entries require a specific kind of structural clarity (the three-reference cluster on USDJPY May 13 is the textbook example) that does not appear often. Maybe one in fifteen setups in our internal data clears the counter-macro path.
What does a 58 percent NEUTRAL confidence read actually mean?
NEUTRAL at 58 percent is the floor for any entry. The Trend Agent will not enter a trade with confidence below 55 percent; the 58 percent reading means the chart cleared the gate but did not produce a high-conviction directional read. The Risk Agent translates that into the smallest position size the scalar table allows: 0.75 percent equity under TRANSITIONING. Higher-confidence reads (BEARISH or BULLISH at 65 percent or higher) get larger size.
When does a USDJPY long get vetoed by the support cluster failing?
When price closes back below the cluster on a confirming candle. The 157.742 stop was set below the entire three-reference cluster plus a 7.5 pip buffer. A close below that level would invalidate all three references simultaneously (VWAP, Fib retracement, EMA support), and the system would exit. The Risk Agent does not move the stop wider on retest attempts; the stop is structural and stays where it was set.
Run your markets with SkyAnalyst
Seven-day free trial. No credit card. Full access to the Trend Agent, Macro Agent, and six-factor confluence scoring.
Trading involves substantial risk of loss. Past performance is not indicative of future results. The analysis shown was produced by an AI model operating on SkyAnalyst’s live trading infrastructure; it is shared for educational and research purposes only and is not financial advice. About reported results. Every AI Trader publishes three take-profit targets (TP1, TP2, TP3) per trade. The broker closes 100% of the position at TP1, so two distinct R-multiples appear in this article. The hero R-multiple is the full-potential R: where the market actually traveled (the highest take-profit hit, or the stop loss) before the setup was invalidated or exhausted. The realized R, shown on the TP1 row of the simulated returns panel, is TP1’s R (or -1R on a stop out). The realized R is what we log to our running track record. Both numbers are honest. Showing both is what lets readers see the full arc of the move and the conservative ledger entry it produced. Simulated returns in this article are calculated against a hypothetical $100,000 account at 2% risk per trade (1R = $2,000). These are educational reference figures and do not reflect any specific account or broker execution. Your actual result depends on your position size, your risk parameters, and live market conditions.
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.