SkyAnalyst AI journal entry: XAUUSD Short on Mar 24, 2026 closed +0.55R on TP1. Full workspace view, decision log, and AI reasoning, unedited.

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.
Gold opened the NY AM session of March 24 inside a clean primary-driver headwind. The 10-year yield was 4.384, above its 5-day EMA of 4.333 and above the recent 3-day sequence trend. The Dollar Index was 99.315, firm above the prior day's 99.163 close. Both readings, taken together, are the dominant macro pair the Macro Agent uses to grade gold's primary backdrop. Both were tilted bearish for gold. The VIX was 26.51, an elevated print that can intermittently support safe-haven flows, but the Macro Agent treated it as a noise factor rather than a thesis change. Breadth read negative at NYAD -302. Oil was mixed to soft against its 5-day EMA. Net read: bearish gold, but with the explicit confidence caveat the Macro Agent keeps in writing for every regime call.
The confidence caveat is the part that did most of the work on March 24. The Macro Agent's bias was lean-bear at 16% confidence with tradeability scored 42 out of 100. That is a low-conviction macro signal by the agent's own scoring, and the system's rule is unambiguous in that case: every setup confidence score gets reduced by one tier before the Risk Agent considers entry. A high-conviction setup becomes medium. A medium setup becomes medium-low. A medium-low setup gets declined. The Trend Agent's read on the same data was BEARISH at 66% confidence, regime TRANSITIONING, with a REDUCE_SIZE recommendation flagged for any entry that cleared the threshold.
That combination, a bearish macro with low tradeability and a bearish trend in a transitioning regime, is the textbook environment for the system's preferred fade-the-rebound playbook. Sell rallies into resistance, not chase strength. The technical picture matched. NY had reversed off the early flush to the London low at 4351.7, reclaimed the 5-minute VWAP at 4392.5, and was pushing into the exact band the Trend Agent had pre-flagged as overhead supply. The 60-minute structure remained bearish with price still below the slow EMA. The 15-minute had rallied into a fib retracement pocket from the 4441.96 to 4388.46 swing: 61.8% at 4408.9, 50% at 4415.2, 38.2% at 4421.5. Three confluent fib levels stacked inside thirteen points, with today's high at 4441.96 capping the structure another twenty points above. That is the resistance cluster the Trend Agent flagged, and it is the cluster the next nine minutes of decision logs were scored against.
The setup the Trend Agent flagged was a Short Rejection from the 4410 to 4421.5 Fib Resistance Pocket. It is a counter-rally fade inside a broader bearish daily structure, gated by a low-tradeability macro and run with a one-tier confidence reduction. Walking through it explains why the entry took five evaluations to clear, why the grade settled at B rather than something higher, and why the Risk Agent took the entry at REDUCE_SIZE rather than the standard 1% risk allocation.
Price establishes a bearish daily bias on the 60-minute timeframe: structure trading at or below the EMA stack, prior session structure broken to the downside, with a fib retracement pocket from the most recent swing acting as the next decision zone. From that posture, the trader watches for a counter-trend rally into a multi-confluence fib resistance pocket. The pocket is what defines the setup. A single fib level is a guess. Three fib levels stacked inside thirteen points, with a higher structural cap behind them, is a structural wall. The entry is not the touch of the pocket. It is the 5-minute bearish reaction inside it: a rejection candle with upper wick larger than body, RSI rolling back below the recent intraday peak, and ideally a 5-minute close back below the lower edge of the pocket on volume.
This is a staple of fade-the-rebound trading inside a bearish daily structure. The math favors the confirmed rejection over chasing the gap-down. Shorting after the gap has already filled half its range exposes the position to the next mean-reversion bar. Shorting near the top of the fib pocket, with a stop placed just above the structural invalidation level, places the entry near the top of the next probable leg lower with meaningful buffer above the highest reference. The R per unit of risk on the cluster fade is structurally better than the R per unit of risk on chasing the breakdown.
The tell on a fade is what the rebound does at the cluster. A rebound that arrives with deteriorating volume, prints an indecision body inside the zone, and gets a 5-minute close back below the lower edge of the pocket is a level holding. A rebound that punches through the upper edge with above-average volume and closes above the highest reference is a level breaking. Without the rejection candle, the pattern is noise. With it, the pattern is signal.
Fib resistance pockets exist because the prior decline left resting offers behind. The first revisit tests whether those offers are still being defended. A bearish rejection with volume confirms the offers are present. The remaining supply at the pocket is structural, and the next leg lower is more probable than the prior rebound was at extension. The 4410 to 4421.5 pocket on March 24 carried three confluent fib references plus the 4441.96 daily high acting as a structural cap. That is a real wall, sized for the move that had just happened.
It fails in the wrong regime, like every fade. A Short Rejection at Resistance inside a confirmed strong-bull macro, or against a tape that is actively flipping risk-on, will see the rebound reclaim the pocket and continue higher. The Macro Agent's regime grade gates the pattern. On March 24 the macro was lean-bear with low tradeability, which is the borderline gating condition: bearish enough to allow the fade, but weak enough in conviction to require a one-tier confidence reduction and a REDUCE_SIZE recommendation on the entry. That is why the grade settled at B rather than higher.
SkyAnalyst does not favor the Short Rejection at Resistance as a strategy. The same week the Trend Agent was running this gold fade, the same agent had run a tactical pullback continuation long on the Nasdaq the day before, a structural pullback long on the S&P the day before that, and would run VWAP rejection shorts on the Nasdaq and S&P the next day. Different instruments, different directions, different regime reads, different playbooks, all running off the same shared Macro and Cross-Asset state objects.
The system reads the tape first and fits the pattern to what is actually there on each instrument independently. It does not show up to the chart with a directional bias and look for opportunities to express it. The four agents running in parallel, trend, macro, cross-asset, risk, each contribute a different lens on what kind of market each instrument is in right now. On gold on March 24, the read was a bearish macro with weak conviction running into a fib resistance cluster on a counter-trend rebound. On other instruments the same week, the read produced longs in the morning and shorts in the afternoon. Same agents, same coordination, different positions. That is the dynamic strategy reading the dogma never gets right, and it is why the system trades each instrument on its own confluence math rather than a unified directional thesis.

Gold is in a macro-led, technically bearish but intraday rebounding environment. The dominant driver pair is still unfavorable for gold: US 10Y yields are rising over the 5-day lookback and DXY is modestly strengthening vs recent closes/above key recent lows, which keeps the primary macro backdrop tilted against sustained upside. That aligns with the Macro Analysis Agent’s lean_bear bias, driven by hawkish Fed repricing and elevated real yields. However, this macro signal has very low confidence (16%) and low tradeability, so all setup confidence must be reduced one tier.
On the technical side, the Trend Agent is BEARISH with 66% confidence, but in a transitioning regime and explicitly says rallies into 4410–4435/4442 are likely supply unless price cleanly breaks 4442. NY has so far reversed off the early flush to the London low (4351.7) and reclaimed 5-min VWAP, which does carry a London trap reversal flavor intraday. But that rebound is currently pushing into the exact resistance band where the broader bearish trend is expected to defend. So this is not a clean trend-day environment; it is a countertrend recovery inside a broader bearish structure.
Because macro tradeability is low, VIX is elevated, and price is approaching overhead supply into the NY AM window, the best opportunities are conditional, not aggressive market entries.
Directional Bias: Bearish
Volatility: High
No halt condition triggered.
Because yields are rising + DXY is firm/strengthening, the dominant pair gives a:
Macro Backdrop: Bearish gold
That remains the primary filter, even with elevated VIX providing occasional safe-haven bounce support.
Both agents lean bearish, but:
So the strongest directional idea remains sell resistance, not chase strength. Per your rule, all setup confidence is reduced by one tier due to low macro tradeability.
These are the critical NY AM decision points.
Only one setup cleanly meets threshold now. A second setup is conditional and only valid after the London Fix window.
Setup #1: XAUUSD SHORT
Need one of the following inside the entry zone:
Confluences:
Risks:
Setup #2: XAUUSD LONG (conditional breakout only, after 10:40 ET)
All required:
Confluences:
Risks:
Meets 5 of 6:
Would normally be High-ish, but reduced to Medium due to:
Meets 4 of 6 only if breakout confirms:
Then No trade.
If you want, I can convert this into a strict execution playbook for the next 60 minutes with “trigger / cancel / no-trade” conditions only.
14:33 UTC, 75% confidence. Price has rallied 58 points off the London low at 4351.7 and is pressing into the 4408.9 to 4421.5 fib retracement pocket from the 4441.96 to 4388.46 swing. Three confluent fib references stack inside the pocket: 61.8% at 4408.9, 50% at 4415.2, 38.2% at 4421.5. Today's high at 4441.96 caps the structure another twenty points above and is also the structural invalidation level for any short. The 60-minute structure is bearish with price below the slow EMA. The Macro Agent reads lean-bear at low tradeability. Both agents lean bearish. Confluence math is at 75% on a B grade. But the trigger I require is a confirmed 5-minute bearish rejection inside the pocket: upper wick larger than body, ideally a close back below 4408.9 on volume. The 5-minute candle in motion is still building a body, not yet a wick. Confidence is high on the structural read, but the specific confirming evidence has not printed. Declining this evaluation.
14:34 UTC, 66% confidence. One minute later and confidence drops nine points because the immediate context is leaning more bullish. Price has held inside the pocket and the 5-minute that just closed printed a small body with a slight upper wick, more indecision than absorption. Volume on the approach is at the 60-period average, neither confirming a stall nor confirming continuation. The structural premise has not changed since one minute ago, the cluster math is still intact, but the rejection signature I require has not printed. The pattern is intact, the trigger is missing. Declining.
14:35 UTC, 61% confidence. Confidence drops another five points. Price has tested the upper portion of the pocket near 4415, the 50% fib, and the 5-minute body printed slightly bullish with the wick on the underside. That is the wrong-direction signature for a fade entry. RSI on 5-minute has lifted. The bounce momentum that carried price off the London low is still intact. The cluster is being tested but not defended. The longer the price holds inside the pocket without the rejection candle printing, the higher the probability the rebound resolves with a break above 4421.5 rather than a reversal below 4408.9. Declining this evaluation.
14:37 UTC, 62% confidence. Two minutes later and confidence ticks back up by a single point. The 5-minute candle that just closed printed at 4413, fractionally below the 50% fib at 4415.2, with a body that closed in the bottom third and an upper wick reaching 4419. This is the early structural pattern I have been waiting for. The wick is meaningful but the close is at the very edge of the pocket, not decisively below the lower 4408.9 fib. Volume on this bar was at the 60-period average rather than above it. The rejection signature is forming. The volume confirmation is still absent. The pattern's discipline is to wait for both confirmations inside the same window. Declining.
14:39 UTC, 72% confidence. The 14:38 5-minute bar closed at 4406, decisively below the 4408.9 lower edge of the fib pocket, printing a bearish body with the upper wick reaching 4418 and volume meaningfully above the 60-period average. Both confirming components, the close below structural support and the volume confirmation, printed inside the same 5-minute window. RSI rolled back below 60 on this bar. MACD histogram turned negative. Cross-asset confirmation: DXY held firm near 99.32 and yields stayed above the 5-day EMA, both still tilted against gold. The structural premise has not changed since six minutes ago. What changed is that every required confirmation finally printed inside the same bar. Confluence math returned 72% on a B grade, above the entry floor. Reducing position size per the REDUCE_SIZE flag and the macro tradeability tier reduction. Entering short at 4410.06, stop 4449, TP1 4388.5.
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 | +0.55R | +$1,100 |
| TP2 hit — not tracked | +0R | +$0 |
| TP3 hit (max potential) — not tracked | +0R | +$0 |
The five-evaluation arc between 14:33 and 14:39 is the structural lesson. A discretionary trader watching the same six-minute window would have shorted the first test of 4410 at 14:33 when the macro was already lean-bear and price was at the densest fib resistance pocket on the chart. The first evaluation saw the touch without the rejection. Evaluations two and three saw confidence drop as the rebound momentum continued to press the pocket without resolving. Confidence dropped to its session low of 61% at 14:35 because the longer price held inside the pocket without rejecting, the higher the probability of an upside break rather than a downside fade. Evaluation four caught the early rejection signature but lacked the volume confirmation. The 14:38 bar finally printed both confirming components inside the same window. The system is not waiting for confidence. It is waiting for the specific confirming evidence inside the same bar.
The second structural lesson is the sizing arithmetic. Confluence math returned 72% on a B grade at the 14:39 entry, which on a clean macro tape would have cleared full 1% risk. On March 24 the macro was lean-bear at 16% confidence with tradeability scored 42 out of 100. The system's rule reduced the setup confidence by one tier and the Trend Agent's REDUCE_SIZE flag instructed the Risk Agent to scale the position down further. The trade entered short at 4410.06 with stop at 4449 and TP1 at 4388.5. That is a stop distance of 38.94 points and a TP1 distance of 21.56 points, a structural 0.55R first target by design. The position closed at TP1 five hours and forty-four minutes later for +0.55R (TP1) and +$1,100 (TP1) on the hypothetical $100,000 account at 2% risk per trade. The reduced-size entry is what made the win meaningful in a regime that explicitly did not justify standard sizing.
A reduced-size B-grade short into a low-tradeability macro and a transitioning regime is exactly the trade most discretionary readers would have skipped. The system took it because the structural arithmetic cleared the threshold, not because the conviction story was clean. - From the desk - March 25, 2026
The March month-to-date tally entering this trade was −3.46R across 34 trades at a 26.5% win rate. March was a difficult month for the system, and adding a +0.55R (TP1) win on a B-grade short into a low-tradeability macro is precisely the kind of trade that keeps the rolling expectancy from bleeding further. The trade is not flashy. It is the median trade. A reduced-size entry, a one-tier confidence haircut, a 21-point first target, and five hours and forty-four minutes of patience between entry and TP1. The next case study in the same week sits in the March 23 weekly recap, and the broader month context is documented in the March monthly recap.
The interesting thing about this trade is what the macro confidence tier reduction did to the entry decision. A discretionary trader reading the same data would have seen the bearish macro alignment and the fib resistance cluster and very likely shorted the first test at 14:33. The picture looked clean enough on the surface. Two of the four required confluences were already in place: macro bias aligned, technical structure aligned. A third, price at a key level, was about to print. The tradeability score, the regime classification, the macro confidence percentile, those are not numbers a discretionary trader keeps in front of them on every chart click.
The system does. The Macro Agent had written its lean-bear bias with the explicit 16% confidence and 42 out of 100 tradeability score into the shared state at 14:00 UTC. The Trend Agent's five evaluations between 14:33 and 14:39 read that state object on every cycle, applied the one-tier confidence reduction to the setup grade, and used the regime classification of TRANSITIONING to instruct the Risk Agent to apply the REDUCE_SIZE flag on entry. None of that is interpretable from a narrative. It is structured data, written and read by separate agents on every cycle, scored against a fixed rule. The discretionary version of the same call would have over-sized.
A reasonable question by now is whether a retail trader with ChatGPT and a chart open could reproduce this. They cannot, and not because of model quality. On March 24 the Macro Agent had committed its conflicted lean-bear read with the explicit confidence and tradeability scores into the shared state at the start of the session and had not updated it since. The Trend Agent's five evaluations between 14:33 and 14:39 read that same state object, used it to gate the setup grade at B rather than the higher conviction the technical picture alone might have justified, and used it to flag the REDUCE_SIZE recommendation that the Risk Agent then enforced on entry. If the Macro Agent had been chatting in prose about a mixed tape, the Trend Agent would have had to interpret the tone five times across five evaluations on a six-minute window. It does not, so it did not. The coordination between the four agents is the product. That is what a chat interface cannot simulate, and it is what this case study shows in practice on a reduced-size B-grade fade nobody would have called the highlight of the week.
The next case study returns to the broader March arc. We will continue working through the month the same way.
From the SkyAnalyst Team.
The Macro Agent grades every regime call along two independent axes: directional bias and conviction. Lean-bear is the directional read. Tradeability scored at 42 out of 100 is the conviction read, and on March 24 it was below the threshold the system treats as actionable on its own. The system's rule for that combination is to allow setups that align with the directional bias but to reduce every setup confidence score by one tier before the Risk Agent considers entry. That is what kept this short from sizing at full 1% risk and why the Trend Agent's REDUCE_SIZE flag was read on entry.
The structural premise being in place is not the entry trigger. The entry trigger is the specific 5-minute rejection signature inside the fib resistance pocket: a bearish candle with upper wick larger than body, a close back below the lower edge of the pocket at 4408.9, and confirming volume above the 60-period average on the rejection bar. The first four evaluations saw the structure intact but the trigger missing. Confidence dropped to 61% at 14:35 because the longer the rebound held inside the pocket without rejecting, the higher the probability the structure would resolve to the upside. The 14:38 bar finally printed all required components inside the same window.
The rolling tally tracks month-to-date, quarter-to-date, and year-to-date net R alongside trade count and win rate. Entering this trade the March MTD was −3.46R across 34 trades at a 26.5% win rate. The QTD and YTD figures matched at −9.79R across 52 trades at a 25% win rate. March was a difficult month and Q1 was a difficult quarter for the system. Adding a +0.55R (TP1) win on a reduced-size B-grade fade keeps the rolling expectancy from bleeding further and is precisely the kind of median trade the journal aims to publish alongside the larger wins.
The pattern fails when the resistance pocket does not hold. On March 24 the stop was at 4449, eight points above the structural invalidation at the day's high of 4441.96. A 5-minute close above 4449 would have invalidated the structural premise and closed the position at -1R. The system does not adjust the stop based on developing information once the position is open. The stop is the line at which the structural premise is invalid, and the trade closes mechanically at that line. The same arithmetic applies to every fade the system runs against any cluster on any instrument.
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. Each model outputs three take-profit targets (TP1, TP2, TP3) per trade. In live execution, models typically scale out at TP1 for risk management — the broker position records this as a TP1 exit. The R-multiples and dollar returns shown in this article reflect the full potential of the trade: where the market actually traveled to (the highest take-profit hit, or stop loss) before the setup was invalidated or exhausted. This lets readers see the complete arc of each setup, not just where the position was closed. 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.
Ninety-nine trades since launch on Jan 12, 2026. Plus 16.57R net at a 58.6 percent win rate. The headline isn't the number — it's how a desk that opened with three trades in January became a system holding expectancy across four months.

A SHORT at 6596.9 into VWAP and prior-day-low resistance, four waits and one enter at 74 percent confidence, a 3h 55m hold to TP1 for +1.18R inside the worst week of the published record.

A SHORT into the 4618 to 4643 NY rebound resistance, eighteen evaluations before the trigger printed at 66 percent, a 3h 59m ride to TP1 for +1R inside the worst weekly stretch of the published record.