XAUUSD signal: sell the $3,966 bounce, target $3,898

Gold closed Friday at $3,875.44, its lowest since October 2025, and spent Monday getting most of it back. The London morning sold it to $3,891.36; the 16:00 UTC bar printed $3,970.31; at 20:00 UTC it was $3,944.71, +1.79% on the day. Gate.com's feed shows an intraday high of $3,970.27 against a $3,874.16 low (Gate), and Kitco had spot at $3,938.90 at 19:50 UTC, about $6 under the aggregate feeds as usual (Kitco). The band is $3,939–3,947.

The rally came in two jumps, and neither was a gold story. Brent fell 5.9% to $106.75 by late morning in New York on reports that Iranian negotiators were in Doha for a third day of talks with Qatari mediators, and gold rose with both of the session's drops in crude (Ausecure). Cheaper oil trims the inflation case behind December hike pricing, and the dollar fell 0.38% alongside it. FXEmpire's read is that the oil move pulled in short covering and speculative buying rather than new longs (FXEmpire).

What did not change is the reason gold fell. December hike odds are 76%, up from 72% at Friday's close, and 83% for at least one hike by 27 January (Reuters, Ausecure). September CPI printed 3.7% year on year, and Dallas Fed's Lorie Logan said on 30 October that she wants two more 25bp increases. A 1.8% bounce does not touch any of that.

The desk's own Friday short was stopped today

Sunday's plan rested a $3,903 sell limit with a $3,963 stop (that plan). Gold traded $3,903.71 in the 00:00 UTC bar, so the order filled, and the 14:00 UTC bar printed $3,966.22 — through the stop. One R gone. The direction held for nine hours and the level for twelve; it is still a loss, and tonight's grading pass books it on the track record.

Two things follow. First, that plan's invalidation clause — an hourly close above $3,963 — fired before the stop did, at $3,964.58. A $7 buffer is not a buffer; today's stop sits 0.1 × ATR beyond its level rather than 0.1 × a dollar. Second, this is a bounce inside a downtrend until the tape says otherwise: gold is still $4,087 on the 20-day mean of closes, October closed −6.78%, and today's high is a lower high against Friday's $4,027.13.

But today's low, $3,891.36, is a higher low. Lower high, higher low — that is compression, not continuation, and it is why this plan carries a defined way to be wrong at about −0.17R instead of a wide stop and a prayer.

Where the levels actually are

• $4,027.13 — Friday's high, and where the liquidation began. Gold opened that session at $4,023.94 and lost $157 to $3,870.17. Reclaim $4,027 and Friday is a spike, not a trend.
• $3,977 — today's high plus 0.1 × ATR. That is this plan's invalidation, not its stop.
• $3,970.31 — today's print at 16:00 UTC. Price has closed below $3,955 in each of the last two hourly bars, so the bounce is already leaking.
• $3,955.32 — Friday's breakdown shelf. Gold held it for four hours today and gave it back at the 18:00 close, $3,953.66.
• $3,898.36 — today's open, and the shelf the 05:00 to 12:00 UTC bars held before the 13:00 jump.
• $3,870.17 — Friday's low, the one-year low. Below it the desk's touch data is thin: one daily extreme in each $10 bucket between $3,810 and $3,860.
• $3,830 — where FinanceFeeds put the bear reference for this week (FinanceFeeds). It is also 2R from today's entry, which is a coincidence worth taking.

The plan

• Instrument: XAUUSD spot, MT5 gold
• Session: Tuesday 3 November — Monday 23:00 UTC reopen (07:00 SGT) to the 20:00 UTC cut (04:00 SGT Wednesday)
• Bias: lower while $4,027 caps it; two hourly closes above $3,977 end the sequence
• Entry: $3,966 sell limit
• Stop loss: $4,034
• Take profit 1: $3,898 — 1.0R, close half
• Take profit 2: $3,830 — 2.0R, close the rest
• Risk-reward: 1:1 to TP1, 1:2 to TP2
• Do not take it if: spot is above $3,966 when you read this — a sell limit under the market fills as a market order at a price the plan was not written for — or gold gaps above $3,977 at the 23:00 UTC reopen.
• Time stop: flat or unfilled by 20:00 UTC Tuesday

The arithmetic, so it can be checked rather than trusted:

• Stop distance $68 = 0.967 × ATR(14) $70.31, computed on fourteen completed daily bars. The floor is 0.7 × ATR, $49.21. The stop clears it.
• $4,034 sits $6.87 above Friday's $4,027.13 high — a 0.098 × ATR buffer, priced off Investing.com's daily and hourly series and measured against the higher of the two feeds, not the friendlier one.
• Entry is $21.29 above spot, 0.30 × ATR — inside the half-ATR limit, and $4.31 below today's high.
• Break-even win rate: 67% if the runner is scratched at entry after TP1, 57% if TP2 prints half the time, 40% if both targets print every time. None of those clears the 55% line, so the simpler version of this plan is to take the whole position at $3,898 and skip the runner — that needs 50%. The 2R target is upside, not the thesis.
• The stop is wide because the last three were not. The first five signals on this desk used $19–23 stops inside sessions that ranged $52–97, and $5.60 and $0.29 decided two of them. $68 on a $70 ATR is the tightest the current volatility allows, and it is why TP1 is a full ATR from the entry.

Execution: the gold spread runs $0.30–0.50 at the 23:00 reopen and around JOLTS. On a $68 stop, $0.40 is 0.6% of the risk. At 0.10 lots the stop risks $680; size from the stop, not from the target.

Tuesday's print, and the one it is not holding

JOLTS at 15:00 UTC (10:00 ET) is the only scheduled US release inside the session, and the first labour read since September CPI printed 3.7% (fedratecalc). It is not tier one, but it is the last input before Friday's payrolls and it lands five hours before the cut. Expect two or three minutes of wider spread and no entries.

The midterm results begin emerging after 23:00 UTC — three hours after this plan is flat, deliberately. A fiscal surprise is a gap the stop cannot price, and the desk does not hold a $68 stop through an election count.

What makes this wrong

Today's rally had a real cause and it may not be finished. Oil is the transmission: Brent at $106.75 is still high enough to keep the inflation case, and every headline out of Doha that suggests a deal that ends the war takes another $3–5 off crude and lifts gold through the hike-odds channel. A 6% one-day drop in crude tends to front-run itself for a week.

Positioning cuts the other way. Managed money was still net long about 105,600 gold contracts as of 27 October, and Friday's selling was margin calls rather than a change of view — Saxo's Ole Hansen called it "technical selling and liquidity concerns overshadowing its longer-term investment appeal" (Reuters). If the longs are still in, there are more forced sellers under $3,870. If Friday flushed them, $3,870 is a floor.

And BullionVault's sample: gold rose in each of the four 40-session post-midterm windows it measured — +2.8%, +0.5%, +3.9%, +11.9% — a small sample from different regimes, and not a forecast, but enough to be flat into the count rather than long or short through it (BullionVault).

If gold closes two consecutive hours above $3,977, the idea is wrong and the plan is out at roughly −0.17R. If it gaps there, the plan pays the full −1.0R and the October ledger has one more example of what a gap costs.

Sources

• Reuters via MarketScreener — gold rises on softer oil, markets parse Fed signals
• Ausecure — gold up 1.7% to $3,943, silver tops $51, oil falls
• BullionVault — gold rallies nearly $75 after an 11% pre-midterms plunge
• FXEmpire — XAUUSD rebounds as the oil plunge pressures Fed hike expectations
• FinanceFeeds — $3,906 Monday, midterms 3 November, jobs 6 November, bear $3,830
• Gate — gold price today, intraday high and low
• Kitco — live gold spot
• fedratecalc — US economic calendar, November 2026
• Price data: Investing.com XAU/USD daily and hourly series, pulled for this session

Not financial advice. Levels are structural references, not predictions, and a stop is a price at which the idea is wrong — not a guarantee of the fill. Gold has moved 3% or more in a session three times in the last six sessions; size accordingly.

Planning a gold position? The signals board carries every plan the desk publishes, and session levels groups them by the level they were built on.