Strategy Analysis

What an NY AM pullback long on USDJPY is supposed to look like

The pattern is structurally identical to the May 18 setup Claude banked the morning before: USDJPY has the tightest yield correlation of any major FX pair, and pullback longs into VWAP or a broken-resistance flip during a rising-yield session are the textbook continuation entry. The model is buying the rate differential between US Treasuries and JGBs; the technical structure exists to time the entry, not to justify the direction.

The template wants six things to line up. The 10-year yield needs to be rising into its 5-day band and above its 5-day EMA. The dollar index needs to confirm without being stretched. The Tokyo session needs to have set a bullish bias. The 60-minute EMA stack on USDJPY needs to be bullishly aligned. The lower-timeframe entry needs to come from a pullback into VWAP or a moving-average cluster or a broken-resistance flip that coincides with one of those higher-frame anchors — and the trigger candle, on the 5-minute, needs to be a bullish reaction print (hammer or engulfing) inside the zone. No high-impact US data event can be scheduled inside the next thirty minutes.

The failure mode the template was built to avoid is the one this trade ran into: entering after the impulse leg has already extended, with no clean pullback to a structural anchor and no clear reversal candle to confirm the retest. The pre-trade analysis was explicit about it. The Confluence Gate output read 7/7 on macro alignment, but the entry section carried a deliberate caveat — "Do NOT chase at 159.20+ — wait for pullback to entry zone. If price runs to 159.30+ without pullback, stand down and reassess."

How Claude structured the entry

The analysis was thorough. The evaluations were patient. The execution was not.

Claude's pre-trade analysis ran 5,502 characters and walked the full template. The 10Y at 4.683% above its 5-day EMA with fresh upside momentum — the carry tailwind was on at its strongest read of the session. DXY at 99.422 above its 5-day EMA and near multi-day highs. The Trend Authority Agent registered USDJPY at BULLISH 65% with a TRENDING regime, support at 158.907, resistance at 159.191, and an invalidation line at 158.907. The Macro Analysis Agent leaned bull at 82% confidence with a tradeability score of 75/100. Tokyo had set its high at 159.028 and price was now trading above it. The 60-minute EMA stack was bullishly aligned. Six of seven confluences cleared cleanly; the seventh — the absence of a high-impact USD event — cleared by default because Pending Home Sales had already released earlier in the day.

The trade plan was specific. Entry zone 159.09–159.17. Stop at 158.88 — below the 15-minute ATR buffer at 159.04 and below the Trend Agent's 158.907 invalidation line. TP1 at 159.244 (the NY/London high, ~1.2R from mid-zone). TP2 at 159.40 (round-level + measured move, ~2.1R). TP3 at 159.55–159.60 (~3R), conditional on the 10Y holding above 4.65% and DXY staying bid. R:R to TP2 was 2.1:1. The confidence rating: 8.5/10 — Very High.

Then the evaluations started, and the trade plan started to drift.

Claude ran seven evaluations in nine minutes, from 14:33:55 to 14:42:49 UTC. The first six all returned wait. The reasoning was consistent and correct each time: price was inside the zone at the upper edge (90–96% position), but the 5-minute trigger candle had not formed. The required bullish reaction (hammer or engulfing) was absent. The 5-minute MACD histogram was weakening rather than stabilizing. RSI was fading from 62 down through 57, 56, 55, 54. Each evaluation noted the same structural problem — "price already tagged TP1 area (159.244) before pulling back, so the impulse leg may be exhausted." The model was reading its own setup honestly. Six times.

On the seventh evaluation, at 14:42:49 UTC, the decision flipped. Claude entered at 159.147 with 62% confidence — a notable drop from the pre-trade analysis's 8.5/10 (Very High) framework grade. The reasoning cited a bullish forming candle "showing reaction off the 159.149 low," RSI at 57 holding above 45, MACD remaining above zero, price holding above VWAP at 159.096, and the Trend Agent's 72% bullish trending regime. The same evaluation acknowledged the offsetting concerns: "MACD histogram weakening and volume is low, plus price is near the top of the zone reducing R:R." The trigger had marginally fired. The framework signed off.

That marginal trigger is the trade. Six evaluations had said the impulse was spent and the reaction candle was missing. The seventh said one forming candle was enough — at 88% zone position with a weakening histogram and declining volume. The 23-percentage-point gap between the framework's pre-trade confidence (85%) and the entry confidence (62%) reflects the same uncertainty the wait-decisions had been recording. The model entered anyway.

What happened after the entry

The trade lasted one hour, thirty-one minutes.

USDJPY pushed marginally higher off the entry, made one attempt at 159.18, and then began grinding lower as the NY AM tape cooled. The 159.149 low Claude had identified as the reaction point did not hold. Price faded through 159.10, paused briefly at 159.05, and traded down into the 158.95–158.90 zone over the next forty-five minutes. The 158.88 stop — placed below the 15-minute ATR buffer and the Trend Agent's invalidation line — was filled at 16:14:17 UTC. The realized R was -1.0R (SL) and the dollar P&L was -$1,000.00 (SL) on the standard $1,000 risk unit.

The structural stop did its job. It sat below the level the Trend Agent had flagged as invalidation; when price reached that level, the bullish thesis was no longer valid and the position closed at the planned exit. The trade did not blow through the stop, did not require manual intervention, and did not produce a larger loss than the framework's risk parameters allowed. That is the entire point of placing structural stops in the first place. The trade was sized to fail at exactly -1.0R (SL), and it did.

The harder question is whether the trade should have happened at all.

Why the second look failed where the first one succeeded

The May 18 USDJPY long entered at 158.824 — the lower edge of a 158.80–158.85 pullback zone — on a single evaluation that registered should_enter: true at 66% confidence. Price had faded from Tokyo's 159.06 high into VWAP, the 5-minute trigger candle had formed, and the impulse leg was still ahead. The trade held for twenty-one hours, paid the structural stop's worth of patience, and exited at TP2 for +$940 (TP2).

The May 19 USDJPY long entered at 159.147 — the upper edge of a 159.09–159.17 pullback zone — on the seventh of seven evaluations, six of which had said the trigger was not there. Price had already extended into the London high zone and was consolidating with a weakening MACD and a declining RSI. The impulse leg was behind the entry, not ahead of it. The macro read was identical to the day before. The execution context was the inverse.

The framework's six-piece template includes the trigger requirement for exactly this reason. A clean macro read does not justify entry in the absence of a clean trigger; the trigger is the part that times the trade so the structural risk lines up against an actual reaction off a defended level rather than against the back end of an already-exhausted move. On May 18, the template ran end-to-end and the trade paid. On May 19, the template's macro layer was strong, the trigger layer was marginal, and the model committed on the seventh look.

The result is not surprising. It is consistent with what the framework would predict for a pullback long entered after the impulse has run, into a zone the model itself was unwilling to commit to on the first six evaluations. The yield read was right. The DXY read was right. The Trend Agent read was right. The trigger was off, and the trigger is the part of the template that controls execution price — which is the part that decides whether the structural stop becomes a -1R fill or a stop that never gets tested.

What this means for Season 2 so far

Claude's first four Season 2 trades sit at 2W-2L with the broker balance at $49,775. The two winners are the USDJPY TP2 (+$940 from May 18) and a NAS100 short that ran to TP2 the same day (+$835 / +0.83R from May 18). The two losers are the US30 long that stopped within two minutes of entry on May 18 (-$1,000) and this USDJPY long (-$1,000). The net P&L across the four trades is -$225 on $50,000 starting capital — a -0.45% drawdown. The season is young.

The pattern worth tracking is the gap between the framework's analytical confidence and the model's entry confidence under marginal trigger conditions. On the May 18 USDJPY entry, the framework graded the setup at 7.5/10 (Medium-High) and the model entered on a single evaluation at 66% — those numbers were closer than usual, and the trade worked. On the May 19 entry, the framework graded 8.5/10 (Very High) and the model entered on the seventh evaluation at 62% — that gap was the largest of the four Season 2 entries so far, and the trade lost. One observation is not a pattern. But the size of that confidence gap is the variable that distinguishes the day Claude rode a clean carry trade from the day it chased one.