READING THE MARKET DEEPERINTERMEDIATE · LESSON 18 / 24~6 min read

Sweep detection + stop-hunt patterns.

A watering hole on the savanna is where the prey gathers and where the predators wait. The same shape exists on every daily chart: clustered stops just above prior highs and just below prior lows are the prey, and the algorithms that hunt them are the predators. A "sweep" is the print where price briefly spikes through a known stop cluster, triggers the stops, then reverses. The framework's sweep_clean field is the binary read of "did a sweep just happen on this name." Tradeable setups change character entirely depending on where they sit relative to a recent sweep — sometimes the sweep is the entry signal; sometimes it's the reason to refuse entirely.

What a sweep looks like mechanically

Three-bar pattern, daily timeframe:

Day N-1 prints a swing low at $L (or swing high at $H). Stops cluster just below (or above) at $L − a few cents (or $H + a few cents).
Day N spikes intraday through the stop cluster — often opening normally, sliding to print a new daily low (or high) by 0.5-1.5× ATR past the cluster, triggering the stops, then reversing within the same session.
Day N closes back above (or below) the swept level — frequently with a long lower wick (or upper wick), giving the candle a hammer or shooting-star anatomy.

The volume on the wick is the signature. A sweep without volume is just a noisy bar. A sweep on 1.5-2× volume is institutional flow exploiting the stop cluster — buying (or selling) into the panicked stop-out flow at favorable prices. Once filled, they release the pressure and price recovers.

What sweep_clean actually reports

The framework's sweep_clean field is boolean: true means no sweep detected in the recent window (typically 5 sessions), false means at least one sweep print exists in that window. Per the dashboard's pillar chips:

sweep_clean = true · ✓ CLEAN. Recent tape is clean of obvious stop-hunts. The level the framework computes from confluence (Lesson 16) is currently the operative one, not a freshly-swept echo.
sweep_clean = false · ⚠ SWEEP. Recent tape shows a stop-hunt print. The level that just got swept is no longer reliable structure — its stops are gone. The next bounce, if any, has to defend a different level.

When sweep makes a setup MORE tradeable

Counter-intuitively, a recent sweep on the daily tape often improves the next setup's quality. The mechanism: stops at the prior level have been cleared. The next test of the same area faces a thinner stop ladder — fewer forced sellers (or buyers) waiting to push price through it. Setups that print a confluence bounce after a sweep (i.e., 2-5 sessions later, at the same level the sweep tagged) are statistically more durable than first-test bounces. The institutional flow that filled at the sweep print has bid support behind the next test.

Practical pattern: NVDA sweeps a $232 swing low to $230.40 on heavy volume Tuesday, closes back at $234. Wednesday-Thursday consolidate. Friday tests $232-233 again. That retest is the higher-probability setup — the sweep cleared retail stops, the institutional bid is now behind the level. The framework's R:R math reads the post-sweep cluster as still valid; the trap detector doesn't fire; the entry trigger arms.

When sweep makes a setup LESS tradeable

The opposite case: when the sweep is happening right now, or within the last 1-2 sessions, the setup is in active stop-hunt territory and the framework refuses entries. The stops haven't fully cleared, the institutional fill hasn't completed, and entering during the sweep typically means buying into the very flow the sweep is exploiting. Pillar chip: ⚠ Sweep detected. New entries refused at preview.

The cleanest tell: the bar is currently printing wide-range with a wick in the wrong direction relative to the prior structure, on volume above 1.5× average. That's the sweep happening live. Wait. The framework's job here is the same as L11's "when NOT to trade" — the cost of waiting one more session is zero; the cost of entering during the sweep is meaningful.

⌬ Sweep aftermath checker

Days since sweep3 days

Sweep volume1.7×

Sweep recovery80%

At confluence?Yes

Sweep qualityHigh (clean stop-hunt)

Aftermath stageCooldown · retest tradeable

VerdictGO — post-sweep retest

Sweep was 3 days ago, volume 1.7× confirms institutional fill, recovery 80% means the panic resolved cleanly. Setup is at confluence (L16). The stops are gone, the level is still defended by post-sweep flow. Entry trigger arms; pillar chips clear.

Drag "days since sweep" to 0-1 — the aftermath is still active, framework refuses. Drag "sweep recovery" below 50% — wasn't really a sweep, just a directional bar. Drag confluence to "no" — sweep was at random level, not tradeable.

Distinguishing sweep from breakdown

Not every print past a prior level is a sweep. A genuine breakdown (or breakout) has different signatures:

Sweep: price spikes through, then reverses same session. Closes back above (or below) the swept level. Long wick. High volume on the wick portion. Volume on the recovery is also healthy.
Breakdown / breakout: price closes through with momentum. No wick — closes near the day's low (for breakdown) or high (for breakout). Volume confirms direction throughout, not just at the wick.

The framework distinguishes these via the close-relative-to-range read. A bar that closes in the bottom 25% of its range (for a low-test) is a breakdown attempt; a bar that closes in the top 75% after testing low is a sweep. The code is mechanical and runs on every audit cycle; the trader rarely needs to second-guess it.

Stop placement around recent sweeps

One operational rule: if you're entering after a recent sweep, your stop goes below the sweep low (for a long), not at the prior swing low. The reason: the prior swing low's stops are gone — they got cleared in the sweep. The new structural low is wherever the sweep wicked to. That's the level next aggressive sellers would have to take out to break the post-sweep structure. Stop placement at the prior swing low (above the sweep wick) gives a tight stop that's structurally weak; stop at the sweep low gives a wider stop that's structurally honest.

This also feeds the L5 position-sizing math: a wider sweep-aware stop means a smaller position. That's fine. The framework would rather have a smaller position with a structurally honest stop than a larger one that's an overnight gap away from getting clipped on the next stop-hunt.

What the framework does

sweep_clean field on every ticker, refreshed each audit cycle
Pillar chip on the audit card showing ✓ CLEAN or ⚠ SWEEP with hover-tooltip explaining the recent print
Refusal on active sweep — entries within 1-2 sessions of a sweep get pillar-rejected
Stop suggestion when entering post-sweep — the chandelier-exit math reads the sweep low as the relevant structural anchor, not the prior swing low

The real lesson

Stops on a chart are prey. Sweeps are how the predators feed. The disciplined trader doesn't try to identify the predators; the framework's job is to identify where the sweep just happened and shift the structural reads accordingly. Setups improve after a clean sweep; setups during one are traps. The sweep_clean field is the binary tell. Combined with the confluence merger (L16) and hidden-tape signals (L17), the three together — clean tape, real confluence, no active sweep — describe the trade-quality floor that the framework's pillar gates enforce.