Gold's London Break Fails 61% of the Time. The Tell Isn't the Poke — It's the Push

Our Asian-close study answered which side of gold's overnight range London runs first. It deliberately left the harder question open: once London takes the level, does the break go anywhere? This study answers that one, and the answer turns out to be more useful than the first.

We took every session from 2 January to 5 August 2026 where the 07:00-09:00 UTC London window broke one extreme of the 00:00-07:00 UTC Asian range, and checked where price sat at the 09:00 checkpoint. 79 breaks.

The break is the losing side

At the 09:00 UTC checkpoint	Count	Share
Back inside the Asian range (break failed)	48	60.8%
Still beyond the broken level (break held)	31	39.2%
Through the opposite extreme	0	0.0%

Three in five London breaks of gold's overnight range are back inside that range within two hours. This is the same personality the opening-range study found on a 16-year sample, measured a different way on a different window, and it lands in the same place: London's first move through a level is more often liquidity being collected than a move being started.

The zero in the last row matters too. Not one of these 79 sessions ran all the way through the far side of the Asian range inside the window. Whatever London does with the level in those two hours, it does it near the level.

The obvious tell is not the tell

The intuitive read — and the one you'll hear repeated as fact — is that a break which barely clears the level was never real, while a break that punches decisively through is the genuine article. We tested it directly by measuring how far past the level price got on the candle that first broke it.

Depth of the first poke past the level	Breaks that failed	Breaks that held
Median	2.58 pts	2.59 pts
Interquartile range	0.71 - 3.97 pts	0.98 - 4.37 pts

Those two distributions are the same distribution. At the moment the level goes, a break destined to reverse and a break destined to run look identical, to within a hundredth of a dollar at the median. Bucketing the first poke by depth produces no usable gradient either: shallow pokes under a point failed 65% of the time, pokes of 1-3 points failed 52%, and pokes deeper than 6 points failed 46%. That is noise around the 61% base rate, not a signal.

If you have been taught to judge a London break by whether it "looked convincing" as it happened, this is the number that should bother you.

What does separate them is extension

Now measure the furthest point the break reached before the checkpoint, rather than where it first crossed:

Maximum extension past the level	Breaks that failed	Breaks that held
Median	3.85 pts	14.80 pts
Interquartile range	2.68 - 7.16 pts	10.14 - 26.27 pts

Those distributions barely overlap. Sorted into buckets the gradient is close to monotonic:

Maximum extension	Sessions	Ended back inside
0-5 pts	29	28 (96.6%)
5-10 pts	19	15 (78.9%)
10-20 pts	21	5 (23.8%)
Beyond 20 pts	10	0 (0.0%)

A break that never manages more than five points of extension almost always ends up back inside the range. A break that manages more than twenty never did, across every occurrence in this window. The cleanest single split sits near ten points: below it, 43 of 48 sessions failed; above it, 26 of 31 held.

Read the two tables together and the mechanism is clear enough. Both kinds of break start the same way, with a small probe past a level everybody has marked. They diverge in what happens over the next few candles: real breaks find continuation flow and keep going, while sweeps run out of buyers almost immediately and get retraced. The information is in the follow-through, not the trigger.

Turning that into something you can act on

An honest caveat first, because it decides how this can be used. Maximum extension is measured across the whole two-hour window, so it is a description of what a failed break looks like in hindsight, not a number you can read off the screen at 07:05. You cannot wait to see the final extension and then act on it.

What you can use is the event that the extension statistic describes: price closing back inside the range it just broke. That close is observable in real time, and it is the moment the break has demonstrably failed rather than merely looked weak. It also carries its own geometry, because a break that failed to extend leaves the level and the wick within a few dollars of each other.

That gives a setup with all three parts defined before entry:

• Trigger. London takes the overnight high or low, and price closes back inside the range.
• Invalidation. Behind the wick of the sweep. On a failing break that is usually only a handful of dollars, which is the entire reason the setup is worth taking.
• First objective. The midpoint of the range price just swept out of, where trapped participants have to buy or sell their position back.
• No trade. Price closes beyond the level and stays there. There is no reclaim, so there is no entry.

One session, drawn out

12 June 2026. The Asian range ran 4,170.56 to 4,230.20, a width of 59.64. At 07:30 UTC London took the low by 0.49 and closed the same candle back inside the range. That close was the trigger at 4,173.65, with invalidation at 4,168.50 behind the 4,170.07 wick and a first objective of 4,200.38, the midpoint. Risk of 5.15 against 26.73, so 5.2R, and the objective was reached inside the hour. Price was 48.93 back inside the range by the checkpoint.

Do not read 5.2R as an expectancy. It is one session, and the multiple is large mainly because the stop was small — the same rule on 15 July, where the low was taken by 1.60 and reclaimed, produces roughly 2R on a wider range. The honest number to hold next to it is the 61% base rate above, which describes how often the setup appears, not how often it wins.

23 July 2026 is the session that keeps the rule honest. The Asian range ran 4,111.92 to 4,140.95. At 07:05 London broke the low by 2.32, which is a slightly deeper first poke than 12 June's 0.49 and looked no different in the moment. Price tried once to get back inside, failed under the line, and extended 25.02 points before closing 18.64 beyond the level. No reclaim, so no entry, so no trade. That session cost nothing to sit out, and the trigger is the only reason it was sat out.

What this study does not tell you

• The 61% is a frequency, not a win rate. It says how often a London break ends up back inside the range. It does not say how often the reclaim entry described above reaches its objective before its invalidation. Those are different measurements and we have not published the second one.
• The window closes at 09:00 UTC. A session that reclaims the range and then rolls over at 10:00 is recorded here as a failed break, because that is what it was at the checkpoint. Reclaims that reverse later in the day sit outside what this sample measured.
• Six months of one regime. The sample runs January to August 2026, over which gold traded from roughly 4,400 down through 4,000 and back. The 16-year opening-range study points the same way, which is reassuring, but 79 events in one regime is a base rate to hold loosely.

What this means

• Judge the break by its follow-through, not its first candle. The probe past the level carries no information at all. Everything that separates a sweep from a breakout happens after.
• A close back inside the range is the tradeable event. It is the moment "this looks weak" becomes "this has failed", and it is the only part of the extension statistic you can observe live.
• The small stop is the point of the setup, not the big multiple. A break that failed to extend puts invalidation a few dollars away by construction. That is where the asymmetry comes from — and it is also why any single result on this setup will look flattering.
• No reclaim, no trade. The trigger's main job is not finding the winners. It is keeping you out of the 39% of breaks that actually run.

Every gold session prep on Open Market Journal marks the overnight range and its midpoint before London opens, and notes whether a break has reclaimed. If you would rather see this applied to live sessions than track it yourself, join the free journal.