Strategy Analysis

What is a pullback short into 5m fib/EMA resistance?

A pullback short is a continuation entry that fires after a corrective bounce inside a higher-timeframe bearish structure. The trade thesis is straightforward: the trend is down, the bounce is corrective, and a structurally-defined resistance band — typically a Fibonacci retracement clustered with a moving-average level — is the highest-asymmetry place to re-engage the dominant direction. The model waits for price to lift into the band, watches for an explicit 5-minute rejection signal, then enters short with the stop just above the next structural level.

The setup resolves on three confirmations. A test of the resistance zone, here 1.16930 to 1.16940 — the 5m Fibonacci 38.2% retracement and the 5m slow EMA confluence. A 5-minute rejection signal: an upper-wick rejection candle, a bearish engulfing close, or a 5-minute close back below the bottom of the zone after testing it. And a higher-timeframe context that has not flipped — DXY still above its 5-day EMA, yields still rising, 60-minute structure still bearish, no high-impact event within 30 minutes. When all three align, the trade is structurally valid. The reward profile runs toward the prior session low; the risk runs to the next intraday resistance.

The honest weakness of the setup is that the trigger is binary. A 5-minute rejection candle gives the model permission to enter — it does not guarantee that the rejection will hold for the next five minutes. In a clean bearish tape with conviction sellers, the rejection compounds: the next candle prints lower lows, the structure validates, the trade works. In a chop-up tape where the bearish structure has already absorbed its directional flow, the rejection is the local high — and the very next candle prints right back through it. The framework cannot distinguish the two cases in real time. The price action is identical at the moment of entry. The difference shows up in the next ten minutes.

How GPT structured the trade

The pre-trade analysis ran 6,325 characters — the framework's full structural read, two qualified setups offered, the higher-quality one taken on the closer pullback. Confluence count: 7 of 8 for the short side. Trend Authority Agent bearish at 75% confidence under a TRENDING regime. DXY at 98.815, above its 5-day EMA at 98.632 and rising. Ten-year yields at 4.394%, above the 5-day EMA at 4.349% and rising. 60-minute EMA structure stacked bearish, RSI sub-50 across the last five hourly candles, MACD histogram negative and expanding. Entry zone at the 5-minute Fib/EMA confluence. 15-minute RSI confirming sub-50 without printing a formal extreme. No high-impact event within 30 minutes of the analysis window — the 10:00 AM ET data print had passed, the 2:00 PM ET FOMC was four hours out.

The single soft confluence sat on Macro Agent confidence at 52% — directionally aligned, just below the 60% threshold the framework treats as a hard-rule trigger. Trend Authority's REDUCE_SIZE flag carried the same regime caution as GPT's earlier XAUUSD short on the same morning — the model's process is to read REDUCE_SIZE as a half-size signal under marginal conditions and as a noted-but-overridden signal when cross-asset confirmation is strong. Wednesday's tape gave it strong cross-asset confirmation. GPT held standard size.

Setup parameters: entry zone 1.16930 to 1.16940, sitting on the 5-minute Fibonacci 38.2% retracement clustered with the 5-minute slow EMA. Stop loss 1.17040, beyond the local pullback shelf and below the Trend invalidation at 1.17113 with a small execution buffer governed by the framework's 1x-60m-ATR floor rule. Take-profits at 1.16830 (the session low and structural support, ~1.0R), 1.16776 (60-minute support and pivot low, ~1.5R), and 1.16730 (the next lower structural support, ~1.9R from the entry midpoint). The risk-to-reward profile was 1.0R / 1.5R / 1.9R — a tighter asymmetry than the framework's typical 1.8R / 3.3R / 5R structural-fade pattern, reflecting the short's location at the beginning of a corrective leg rather than at the end of one.

Risk on the trade as set by the framework: $498.61 against a $49,860.58 starting balance, exactly 1.0%. Position size 9.56 lots. Standard risk allocation, decisive sizing, no hedging. Same posture that produced the Tuesday EURUSD win, the same posture that took the morning's XAUUSD loss, the same posture that took this one. The framework does not adapt to the previous trade's outcome.

The seven-evaluation entry — patience that didn't pay

GPT's evaluation log on this trade is seven rows over nine minutes. Six waits, then one enter. That is the longest qualifier sequence the model has run on a EURUSD setup in three weeks, and it is worth reading carefully because it shows the framework's trigger-discipline working exactly as designed — and producing exactly the wrong fill anyway.

Evaluation 1 at 14:34 UTC, confidence 83: price entering the short zone, exact 5-minute trigger has not printed yet, no completed rejection candle, no bearish engulfing, no confirmed 5-minute close back below 1.16927. Wait. Evaluation 2 at 14:35 UTC, confidence 82: price inside the zone at 1.16937, broader context still supportive, but the required trigger has not been confirmed. Wait. Evaluation 3 at 14:37 UTC, confidence 86: still inside the zone, last completed candle closed bullish near the highs at 1.16938. Wait. Evaluation 4 at 14:38 UTC, confidence 79: the last 5-minute candle pushed up into the zone and closed bullish at 1.16938. The current candle is forming with no clear upper-wick rejection. Wait. Evaluation 5 at 14:39 UTC, confidence 53: micro confirmation incomplete, this is close but not a valid short entry yet. Wait. Evaluation 6 at 14:41 UTC: AI evaluation unavailable — conservative default fired. Wait.

Evaluation 7 at 14:43 UTC, confidence 82: enter. Reasoning, verbatim: the 5-minute entry trigger appears satisfied, price tested the 1.16930-1.16940 resistance zone with highs to 1.16942-1.16948, the next closed 5-minute candle finished back below 1.16927 at 1.16922, current price at 1.16918 is still close to the zone, structure remains bearish, no SL or TP touched, the pullback entry remains valid.

That is the framework working as the spec describes it. The model held position discipline through a five-minute window where the trigger almost printed and didn't. It only entered when the conditional sequence completed. By every process metric available, this was a textbook qualifier-and-enter sequence. The complaint — and there is one — is that the rejection candle that finally fired was a local high, not a structural high. The next five-minute candle printed at 1.16930, the one after that at 1.16942, the one after that at 1.16955. Four candles after entry, price was eight pips above the entry. Twelve candles after entry, the stop hit.

The stop loss

Entry 1.16933 at 14:43:18 UTC. Stop loss 1.17040, 107 pips above entry on a $1,074 risk allocation. Take-profit-1 at 1.16830, 103 pips below entry. The trade needed price to break the 1.16880 to 1.16940 chop band to the downside within 30 to 60 minutes to remain viable.

It did the opposite. Price held inside the chop band for forty minutes, then lifted through 1.16950 on the 5-minute, then through 1.16980 on the next, then ran the 1.17000 round number on what looked like a thin-volume squeeze rather than a directional thrust. The 1.17020 print came inside the next ten minutes. The broker stop hit at 1.17039, a single pip inside the analysis stop at 1.17040. Net loss: -$1,080.28 (SL). R-multiple on the broker outcome: -1.0R (SL). Account balance after: $48,780.30. Duration: 3 hours 40 minutes from entry to exit.

Read the stop placement first. The hard stop at 1.17040 was structural — beyond the pullback shelf and below the Trend invalidation at 1.17113, with a small slippage buffer that satisfied the framework's volatility floor. The actual exit at 1.17039 came in clean, one pip inside the analysis stop, no slippage. This was the structural stop firing exactly where the framework specified. The dollar magnitude — -$1,080.28 against a framework-set risk amount of $498.61 — reflects the broker position's full lot exposure rather than the analysis-side risk envelope: the framework sizes for 1% of starting equity at the structural R, but a full broker stop on 9.56 lots at 107 pips of adverse move books the larger figure. -1.0R is the right read against the framework's structural target distance. The dollar number is what the broker position actually paid.

The reversal that came after

This is the part that requires honest reporting. Within the same hour that the broker stop fired, EURUSD reversed direction. The 5-minute candle that printed the 1.17039 high closed back below 1.17030. The next closed below 1.17000. By the close of the next hour, price had returned to 1.16940 — back inside the original entry zone, on the wrong side of the broker exit. By the end of the New York session, EURUSD had stair-stepped down to 1.16830 — the analysis TP1. By the close of Asian trading, it had reached 1.16776 — the analysis TP2. The 1.16730 analysis TP3 hit during the Sydney session.

The analysis-side trade was right. The model identified a corrective bounce, sized a structural short against it, set targets at three structural-support levels, and watched the market deliver to all three. Hours later. Without the position. The enriched trade record initially logged this as a tp3_hit at +1.21R because the price did, eventually, reach TP3. The broker record logged it as sl_hit at -1.0R because the broker had already exited.

This article reports the broker outcome. The dollars are dollars. The position closed at -$1,080.28. The eventual TP3 happened to a chart, not to GPT's account. Anyone trading the broker side of this setup booked the loss at -1R; the chart's later behavior was informative but not financial. The framework's job is to size the trade and place the stop. The market's job is to either pay or not. On this trade, the market shook GPT off the position first, then paid the position GPT no longer held.

The same setup, two hours later, taken by Claude

The narrative parallel is sitting one trade away. Claude entered the same EURUSD short two hours later, at 1.16876 — fifty-seven pips below GPT's entry, after the squeeze that ran the GPT stop had exhausted itself. Claude's stop at 1.17040 was the same level GPT had used. Claude's targets were the same three structural levels. The only meaningful difference between the two trades was timing: GPT entered at the start of the corrective squeeze, Claude entered after the squeeze had spent itself.

Claude's trade ran clean to TP3 at 1.16730 over the next four hours. Same instrument. Same direction. Same structural framework. Same stop. Same targets. Different entry timing, different outcome. This is the inverse of Day 10's US30 long, where the earlier entry won and the later entry was already late. On Day 12, the later entry won. The setup framework cannot tell, in the moment, which case is which.

The lesson is not that Claude is the better model. The lesson is that on this specific setup, on this specific tape, the entry trigger fired twice — once early, once late — and only one of them survived contact with the market. Both models read the same structural read. Both models took the same setup. Both arrived at the same target. The market paid one and not the other. That is the cost of trading a system that fires on a moving market.

Why -1.0R is what discipline costs

The framework was built around this number. Standard 1.0% risk per trade means GPT can take ten consecutive structural losses and still hold 90% of starting equity. It can take fifteen and still hold 86%. The math of the benchmark assumes some setups will fire correctly into wrong-side timing, and the only thing that determines long-run outcomes is whether the model holds the same posture across them.

GPT held the posture today on both losses. Standard 1.0% risk, structural stop, qualifier-then-enter sequence on the EURUSD setup, single-evaluation entry on the XAUUSD setup. No revenge trade after the morning's loss. No size adjustment after the squeeze took the EURUSD short out. No second EURUSD attempt to "get the trade back." The next setup will use the same risk allocation against a different structural read, and the model will not remember today happened. That is not stoicism. It is what the framework requires.

The complaint about Wednesday is not the dollar loss. It is the regime read. Day 11's recovery rhythm — sell rallies, hold for the structural target, let Asia walk price down — produced three consecutive winning sessions before this one. Day 12's tape gave that exact framework two structurally-clean setups that did not work in real time and one that did, two hours later, when Claude took it. The recovery was not a structural lead. It was three favorable tapes, and Wednesday is the first session that breaks the run.

What this trade says about the process

Three reads.

First, the trigger discipline held. Seven evaluations, six patience, one enter — when the conditional sequence completed. The framework specified a 5-minute rejection close back below the bottom of the zone after testing it. The trigger printed at evaluation 7. GPT entered. The model did exactly what it was supposed to do, and the trade still lost. That is not a process failure. That is the framework absorbing a wrong-side fill inside its risk envelope.

Second, the broker-versus-analysis split is the honest signal. The analysis side of the trade reached TP3 at +1.9R hours after the position had been stopped. Reporting this as a +1.9R outcome would be data misuse — the dollars never lived in the account. Reporting it as -1.0R is the only honest read. The framework's value is in the entry framework, the structural targets, the risk envelope. The execution timing is real. The execution timing was the variable that broke today.

Third, the cross-model parallel is the real signal. Claude took the same setup at a better entry. The setup was correct. The implementation was a timing function. On Wednesday, the timing function went against GPT and for Claude. The leaderboard rotation that closes Day 12 — Claude's first lead since Day 7, GPT's two-day giveback of Tuesday's recovery — is not a process gap. It is a tape that paid one model's timing and not the other's.

The market struck Wednesday. -1.0R (SL) is the cost. The model takes the next trade.