The Stop-Hunt Pattern Is Real. The Edge Is Not.

Scope: measured on SPY only, against four levels fixed before the open (prior-day and overnight high and low). Intraday swing levels and round numbers are untested, and these are unconditional averages.

Price broke a pre-listed level and traded back through it within the hour in 73.7% of the 482 breaks we measured — the pattern every stop-hunt story is built on. But any comparable move re-crosses the price it just went through 74.3% of the time. The sweep-and-return shape is not evidence of anyone hunting anyone: it is what price does behind every move, level or no level. We also tested the version sweep traders actually trade — a wick through the level that closes back inside — and fading those paid nothing either.

What counts as a break here?#

Only levels a reader could have written down before the open, fixed in advance as our methodology note specified: the prior day's regular-session high and low, and the overnight high and low (04:00–09:29). A break is the first one-minute close through one of those four prices by at least 2 basis points, between 10:00 and 15:00 ET. No swing points nominated after the fact, no round numbers, no levels chosen because a reversal happened there. SPY, February 2025 through August 2026: 482 first breaks across 309 sessions, split almost evenly — 242 up, 240 down.

Every break is compared against ordinary minutes that were doing something similar: same half-hour of the day, an approach move of the same size once you adjust for how jumpy that day was, at least 0.15% away from all four levels, and no break within half an hour either side. There are 46,308 such control minutes. Confidence intervals throughout come from a bootstrap clustered by trading day, because breaks on one day are not independent of each other. The question is never "what happened after breaks" — it is "what happened after breaks that would not have happened anyway".

Does breaking a level change what comes next?#

No. Return along the break's direction, minus the matched control's:

Both popular readings fail at once. The breakout reading — momentum through a key level continues — shows continuation at or below the ordinary-minute baseline and zero excess drift. The stop-hunt reading — the break is a trap, fade it — needs negative drift, and there is none: the interval excludes any effect larger than about 3 bp per hour in either direction. Broken out by level type, no single type shows an excess its interval can hold, though with 95 to 145 events each those tests are underpowered and should be read as "nothing detectable", not as four-way confirmation.

Then why does everyone remember the sweep that reversed?#

Because the shape is real, common — and universal. Within an hour, 73.7% of level breaks traded back through the broken level. Watch that a few times and "they took the stops and reversed" writes itself.

Now the base rate. Take ordinary moves in the same sessions that pushed through the price just behind them by the same 2 basis points, and ask how often price came back through it — any price, no level involved, no stops resting there:

Every interval contains zero. Price revisits recently crossed prices constantly — that is what a volatile, roughly efficient tape looks like — and doing it at yesterday's high is not detectably different from doing it anywhere else. The level supplies the story, not the behavior.

What about the wick sweep?#

That is the fair objection to everything above, so we tested it. The textbook sweep is not a close through the level — it is a wick through it that closes back inside: price pokes above yesterday's high, takes out the stops, and rejects within the same minute. That is a different event, and one-minute high/low data can express it.

384 such sweeps met the same standard as the breaks — first sweep of each pre-listed level, at least 2 basis points of penetration by the wick, close back on the original side, 10:00 to 15:00 ET. Fading them, which is the trade:

Nothing, at any horizon, with win rates below a coin flip. The canonical sweep is not a weaker version of the effect we failed to find in close-throughs; it is the same absence, measured on the event that the story is actually about.

None of this proves nobody ever runs stops. Intent is invisible in aggregate data, and one-minute bars still cannot see the sequence of fills inside a minute. What it shows is narrower and more useful: whatever happens at these levels leaves no directional trace that separates them from ordinary price, on either the close-through or the wick version — so a strategy keyed to the sweep pattern is trading the base rate and paying costs for it.

What changes tomorrow#

When the next sweep-and-reverse chart crosses your feed, the two questions from the methodology note now have measured answers on SPY. Compared to what? — re-crossing happens behind three-quarters of all moves, so the pattern alone carries no information. Was the event defined before the outcome? — here it was, twice, on both event definitions, and both came back zero. A level break tells you where orders clustered; it does not tell you where price goes next.

Scope: SPY only, four pre-listed level types only — intraday session highs and lows, equal highs and lows, round numbers and weekly levels are untested and may behave differently. First break and first sweep of each level only; repeated tests within a day are not counted. These are unconditional averages: a conditional setup that requires confluence beyond the sweep itself is a different claim and is not tested here, though the burden now sits with it. One-minute bars miss the ordering of trades inside a minute on both the event and control side. Window: 2025-02-10 to 2026-08-28. Not trading advice.

Related reading#

• Before you believe a reversal story — the note that specified this test before any result was known.
• Does price really snap back to VWAP? — the companion measurement; the other reversal story, same verdict.
• What actually moved Bitcoin in August 2026 — the same discipline applied to a macro narrative.