Strategy Analysis

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

USDJPY has the tightest yield correlation of any major FX pair. Long-side setups on the pair are essentially carry trades dressed up in technical language — the model is buying the rate differential between US Treasuries and Japanese government bonds, and the technical structure exists to time the entry rather than to justify the direction. That framing matters for the post-mortem on this trade, because the technical structure on May 18 was good but not pristine, while the yield read was unambiguous.

The template for the NY AM continuation long has six pieces. First, the 10-year yield needs to be rising into its five-day band and printing above its five-day EMA — the carry tailwind has to be on. Second, the dollar index needs to confirm direction without being extended. Third, the Tokyo session needs to have set the high or held a bullish bias into the New York handoff. Fourth, the 60-minute EMA stack on USDJPY needs to be bullishly aligned. Fifth, the lower-timeframe entry needs to come from a pullback into VWAP, a moving-average cluster, or a Fibonacci retracement that coincides with at least one of those higher-frame anchors. Sixth, no high-impact US data event can be scheduled inside the next thirty minutes.

The pattern works because the carry bid is structural and slow, while the technical pullbacks are mechanical and fast. The yield differential is moving over hours and days. The 5-minute candles are moving over minutes. A pullback into VWAP inside a rising-yield day tends to find bid because algorithmic carry desks are doing the same arithmetic the model is. The failure modes are well-documented: a 10Y reversal lower mid-trade, a VIX spike that triggers a carry unwind, or a verbal-intervention headline from the Japanese Ministry of Finance. None of those were active on May 18.

How Claude structured the entry

The analysis ran long. The decision was quick.

Claude's pre-trade analysis on USDJPY was a 10,455-character read that walked the six-piece template end-to-end. The 10Y was at 4.595% with a fresh 5-day high at 4.631% earlier in the session — yield veto for shorts was triggered, and the long-side carry bias was confirmed at the strongest read of the day. DXY at 99.08 was supportive without being stretched. The VIX at 18.42 was modestly above its 5-day EMA but well under the spike threshold. The NYSE advance/decline line was strongly positive at +1,043. Gold was non-divergent, sitting below its own 5-day EMA.

The Trend Authority Agent read USDJPY at BULLISH 62% with a TRANSITIONING regime flag — a moderate-confidence directional read with an explicit "reduce size" suggestion baked in. The Macro Analysis Agent leaned bull at 70% with a tradeability score of 68/100. The Tokyo session had set its high at 159.059, faded into the New York open, and was now consolidating in the 158.80–158.85 band. The 60-minute EMA stack was bullishly aligned. Both London and New York were respecting 158.637 as the structural low. The 5-minute MACD histogram had flipped positive on the latest closed bar and the 5-minute RSI was at 57.9. Six of the seven confluence checks cleared cleanly; the Tokyo bias check was partial because price had faded from the session high rather than holding it.

The confluence score was 6/7. Claude's framework grades that at 7.5 out of 10 confidence — Medium-High. The trigger band was 158.80–158.83. The model committed at 158.824 on a single evaluation that returned should_enter: true at 66% confidence at 14:52:19 UTC. The reasoning was clean: "Price is at 158.829, right at the top of the entry zone (97% position) and just above VWAP (158.822). The 5m trigger conditions are largely met: price reclaimed VWAP, RSI is 58-62, MACD is positive with signal line above zero, EMA fast crossed above slow, and the last 5m candle pulled back from 158.896 to 158.866 — a healthy retest near VWAP/EMA confluence. Trend Agent confirms bullish 64% alignment. Risk is the resistance cluster at 158.88-158.907 just overhead." One evaluation. One entry. The stop went to 158.69, the TP1 at 158.95, the TP2 at 159.06.

What happened after the entry

The trade did not pay quickly. It paid eventually.

The first attempt at TP1 came inside the New York afternoon. Price probed 158.931 — within two pips of the 158.95 target — and rejected at the London high. Claude's own mid-trade self-assessment, written during the position, flagged the failed probe and recommended a manual exit at break-even. "The trade thesis was a NY AM continuation driven by yield/carry momentum extending into London high and Tokyo high. That thesis was tested — price got within 2 pips of TP1 and was rejected." Under the experiment's exit policy, the model's mid-trade discretion does not override the static TP and SL levels. The position stayed open. That decision turned out to be correct.

The New York afternoon ground sideways into a tight range. London closed without price reclaiming the highs. Tokyo opened, and the yield bid that had been the entire premise of the trade reasserted itself overnight. USDJPY moved from 158.84 to 158.96 over the Asian session, hit TP1 at 158.95 cleanly during the Tokyo handoff to London, and continued past 159.00 into the next London cash session. TP2 at 159.06 — the Tokyo high from the prior day, which had been the day-1 ceiling — printed at 11:49:47 UTC on May 19 and the broker filled the close. The realized result on the broker side was +$940 (TP2) at the TP1-distance R-multiple of +0.94R (TP2). Total hold time: 20 hours 57 minutes.

The trade hit both TPs in the order the framework intended, with the structural stop at 158.69 never tested. The mid-trade self-assessment that recommended an early exit was a coin-flip on the day's path; the static-rules policy that kept the position open was the harder discipline. That policy is the reason the realized number is +0.94R (TP2) rather than a small scratch.

Why the yield read was the trade

The technicals on this setup were good but not great. The 6/7 confluence score and the 66% entry confidence reflected that honestly. The Tokyo bias was only partial. The Trend Agent flagged a transitioning regime. The VIX was modestly elevated. A strict-criteria model would have asked for a clean 7/7, waited for the New York session to confirm the trigger with a stronger bullish reversal candle, and would not have committed at 158.824.

What made the trade work was a higher-frame signal Claude's framework explicitly weights as the primary driver: the 10-year yield was rising into a fresh five-day high. That single read overrides a lot of intraday technical noise on USDJPY. The pair's correlation with the 10Y is structural and persistent. When yields are working higher into a session and the dollar is confirming, USDJPY pullbacks into defended VWAP zones are the textbook continuation entry — even when the lower-timeframe trigger is marginal.

This is the inverse of the failure mode on Claude's Season 1 NAS100 stop, where a strong structural read with a weak trigger ran into a market that did not respect the structure. On USDJPY, the structural read was strong (yields rising), the technical read was decent (six confluences), and the trigger was acceptable. The combination produced the outcome the framework is calibrated to produce.

How this fits Season 2 Day 1 for Claude

The season opened on uneven ground. Claude's first trade of the day, a US30 long fired into the New York cash open, stopped within two minutes for -1.0R / -$1,000 (SL) — a position that never had time to develop. The USDJPY entry came forty-seven minutes later, and it was the second trade on the running ledger. Claude entered with the balance at $49,000. The TP2 exit lifted the balance to $49,940 (TP2). A subsequent NAS100 short ran into TP1 territory and was filled at TP2 as well, before three later trades produced losses that brought the running balance to $47,775 after six trades — a -4.5% drawdown on the first lane of trading days.

The USDJPY long is the cleanest piece of Season 2 work Claude has produced so far. The macro read was right. The technical read was honest about its own limitations. The entry was committed without a multi-evaluation grind. The static-rules policy held the position through the failed afternoon probe, and the trade paid on the next session's continuation. The other five trades on the running ledger are a mixed bag. This one is the template the model wants to repeat.

GPT-5.5 traded USDJPY long the same day at a different entry zone with different timing; the two models' results on the pair are part of the head-to-head ledger the Season 2 scorecard updates daily. The yield-driven USDJPY continuation is a setup both models will see often this season. Whether they take it the same way, and what the cumulative outcome looks like, is what the benchmark is measuring.