Support and Resistance Clusters: Wall or Launch Pad? 195 EUR/USD Breaks
Worked (%)
A companion study on this site found that EUR/USD levels only predict follow-through when several independent signals converge on the same price band. That answered which levels matter. It left the harder question open: once a cluster like that breaks, which way does price actually go, and what separates the breaks that run from the ones that trap?
So we went through them one at a time. Every A-grade EUR/USD cluster break over five months, inspected by hand across the daily, four-hour, hourly and one-minute tape, with a written verdict for each: did this run in the direction of the break, or did it trap? 195 breaks.
How we tested it
This is the eyes-on companion to the mechanical study. Each break was reviewed manually rather than scored by a rule, because the first pass showed the automatic labels were unreliable (more on that below). For every break we recorded the higher-timeframe context, how price approached the level, the texture of the break candle, what happened next, when the real move actually fired, and a verdict of worked, trapped, or stand-aside.
Two things follow from that. The counts are approximate, because a handful of cases genuinely sit on the fence. And the verdicts carry analyst judgement, which is a real limitation: this is a structured review, not a mechanical backtest, and no trades were taken. Treat the numbers as the shape of the sample, not as precise probabilities.
Direction decided it, not the level
Sorting the 195 by what kind of setup they turned out to be produces a very wide spread:
| Setup type | Breaks | Worked | Share |
|---|---|---|---|
| Sweep and reclaim (stop-run through the cluster, then reclaim) | ~28 | ~24 | ~86% |
| Pullback into a cluster with the trend | ~34 | ~29 | ~85% |
| Launch shelf (rich cluster acting as a coil in a trend) | ~18 | ~15 | ~83% |
| Fade at an extreme | ~22 | ~13 | ~59% |
| No-trend chop, dead book | ~17 | ~3 | ~18% |
| Break taken against the live thrust | ~45 | ~6 | ~13% |
The top three are the same trade wearing three different hats: all of them go with the recent multi-day direction. The bottom row is the same trade taken backwards. Nothing else in the study came close to this gap, and it held in both directions: the sample splits roughly evenly between up breaks and down breaks, and in down weeks the down breaks worked while the up breaks trapped.
By live thrust we mean the last two or three daily closes together with the four-hour structure, not a moving average. That distinction mattered more than it sounds. Through one extended down week the daily EMA label read "up" for nearly every instance on the tape while price fell the entire time.
A rich cluster in a trend is a coil, not a ceiling
The launch-shelf row is the counter-intuitive one, and it is the reason this study exists.
The instinct when four or five independent signals stack at one price is that the level will hold. In a live trend it usually did the opposite. Price coiled under the band for a few bars, shook through it once, and then expanded away in the direction it had been going all along. In this sample around 15 of 18 of those coils resolved that way. The trades that lost money on those days were the ones that read the confluence as a wall and faded it.
The shakeout is the part that gets people out early. It is common for price to dip through the far side of the cluster first, which reads as a failed breakout right up until the session-open candle closes back through and keeps going.
The move fires at a session open, not at the break stamp
Where the break happened on the clock mattered almost as much as its direction:
| Ignition window | Share of genuine moves |
|---|---|
| London and EU open (about 05:00 to 09:00 UTC) | ~70 to 75% |
| New York and the overlap (about 12:00 to 15:00 UTC) | ~15 to 20% |
| Asian session | ~10% |
A break during the Asian session mostly told us which direction to watch. Acting on it meant sitting through hours of chop before anything happened, and frequently getting whipsawed out first. When the break itself landed inside a live session, the break candle and the real ignition were the same thing.
What the automatic labels got wrong
Worth stating plainly, because these are the fields most retail tooling shows you:
- A big displacement number is not conviction. The loudest break candles in the sample were among the worst trades, because a large move measured against a dead overnight book is an artifact of the thin book, not participation. Absolute volume during a real session separated the genuine breaks. A volume ratio spike against an empty Asian baseline did the opposite.
- The four-hour outcome window understated the good trades badly. It routinely expired during the overnight chop, before the real move started. Winners aligned with a trend day commonly travelled 25 to 90 pips intraday, and the best coil-and-release setups seeded multi-day legs of 130 to 200 pips.
- The "retest held" flag fired on nearly everything, so it carried no information. What discriminated was the quality of the retest: price holding beyond the broken edge versus closing back through it inside the first half hour.
Management followed location
One pattern was consistent enough to write down. Where the setup sat in the range decided whether to hold it or bank it.
Moves that fired from an already-extended extreme snapped back fast and only paid if taken quickly. Moves that fired mid-trend, out of a coil or a sweep-and-reclaim with open road to the next cluster, were the ones that produced the long legs. Those are the cases where a fixed target is the expensive choice, because the distribution of how far they run has a long tail and a fixed cap cuts exactly the trades that pay for the losers.
What this means
- Read the cluster for importance, read the trend for direction. Confluence tells you price will react at that band. It says nothing about which way.
- A multi-signal level inside a live trend is more often a launch pad than a wall. Expect a shake through it before the expansion.
- Wait for a session open. An overnight break is a heads-up, not a trigger.
- Distrust loud break candles and short outcome windows. Both mismeasure what actually happened.
What would change our mind
This is one instrument over five months, reviewed by hand. A quieter, rangebound regime would produce fewer trend days and would likely compress the gap between the with-trend and against-trend rows. The counts are approximate and the verdicts are judgement calls. We are re-running the same inspection on gold and the S&P 500 to see whether the direction-first result is structural or specific to EUR/USD in this window.
This is a historical study of price behaviour, not investment advice, not a signal service, and not a record of trades taken.
Studies like this become the filters inside a documented playbook — the research → playbook → backtest → live loop, locked to one instrument at a time, rather than a scanner firing on everything.