Edges
Judas Swing: How to Tell a Fake-Out From a Real Breakdown
A Judas swing is the opening move that betrays you: price drives hard in one direction, takes out an obvious level, triggers every stop behind it — and then reverses and spends the rest of the session going the other way. The name is the tell. It is the move that looks like the day's trend and is actually the day's trap.
Here is the problem nobody's Judas tutorial admits. At the low, a Judas swing and a genuine breakdown are the same picture. Both sweep a level. Both do it on heavy volume. Both look like capitulation. The difference only appears afterwards, and the thing that reveals it is not a candle shape — it is a reclaim, measured against a clock.
This article gives you that test, checked line by line against a real session's tape, and then does the thing the rest of the internet refuses to do: it tells you honestly how much you do not know about whether this edge exists.
What a Judas swing actually is
A Judas swing is a liquidity raid disguised as a breakout: price sweeps a level where resting stop orders cluster, collects that liquidity, then reverses back through the level and expands in the opposite direction. The move is not information — it is inventory. Someone needed the other side of the trade, and the stops beyond an obvious level are where the other side is parked.
Three ingredients make it, and all three are required:
- An obvious level — premarket low, prior-day low, opening-range low, equal lows. Somewhere every retail chart draws a line and tucks a stop underneath.
- A sweep with force — price trades through it on a volume climax, not a drift.
- A reclaim — price closes back above the swept level, fast, and expands away from it.
Miss ingredient three and you do not have a Judas swing. You have a breakdown. The entire skill is telling those apart in real time, when you do not yet know which one you are looking at.
At the low, the fake-out and the real thing are identical
This is the sentence to keep: you cannot classify a Judas swing at its extreme, because at the extreme it has not happened yet.
Consider what you actually see 25 minutes into a session where price has swept the prior-day low and is grinding lower on huge volume:
- The level is broken. ✓ (a Judas needs this — so does a breakdown)
- Volume is enormous. ✓ (a Judas needs this — so does a breakdown)
- Price is making lower lows. ✓ (a Judas needs this — so does a breakdown)
- It feels like the bottom. ✓ (feelings are not in the dataset)
Every single observable is shared. The bearish case and the bullish case fit the same chart perfectly. If you buy here, you are not reading the market — you are guessing which of two live hypotheses is true, and the reason it feels like reading is that one of them will be right and you will remember the times it was you.
The reclaim test — the one thing that separates them
The discriminator is not the sweep. It is what happens after, and how fast.
The clock does the work. A trapped side is forced — it must buy back, and it must do it now. That urgency is why real raids resolve in minutes. A reclaim that takes 45 minutes is not a squeeze; it is a market slowly changing its mind, which is a different event with different odds.
The mirror image is just as sharp: every minute price stays below the swept level without reclaiming makes a reversal less likely, not more. Time below a level is the market accepting the price. Acceptance is the opposite of a raid. This is the exact point where hope inverts the reasoning — traders treat a longer wait as a bigger coiled spring, when the data-generating process says the opposite.
Run your own sweep through the clock:
What the tape actually did: TSLA, 16 July 2026
Everything below is real, pulled from consolidated (SIP) tape rather than memory. This is the session that prompted this article.
The setup going in:
Price opened at 392.35 — above both levels, with stops resting under each.
What happened, minute by minute:
| Time (ET) | Event | Price |
|---|---|---|
| 09:30 | Opens, sells instantly. Takes premarket low and prior-day low in the first 5-minute bar | 392.35 → 387.20 |
| 09:30–09:50 | Grinds lower. Lower low, lower low, lower low | → 385.70 |
| 09:55 | Session low. Nothing here says reversal | 385.32 |
| 10:00 | Closes back above anchored VWAP. Range 3.31 vs ~1.5 average — displacement | 388.93 |
| 10:05 | Reclaims the prior-day low | 390.83 |
| 10:15 | Structure shift confirmed, higher highs | 393.85 |
| 10:20 | First pullback holds | 391.21 |
| 10:45 | High of day | 395.31 |
The first 5 minutes traded 1.40 million shares — nearly three times the entire premarket session combined. That is the volume climax the definition asks for.
TSLA 16 Jul 2026 — the opening Judas, 5-minute bars (real SIP data)
Now the number that matters most. The low printed at 09:55. Anchored VWAP was reclaimed at 10:00. The reclaim took five minutes — one bar. That is what a forced side looks like, and it is the whole reason this qualifies.
Notice what the tape says about the low itself. At 385.32, there was no reclaim, no displacement, no structure shift — every confirming input was still blank. Buying 385–386 was directionally correct and evidentially empty; the same action on a day that kept going would have been the top of a losing trade. The evidence arrived 10 to 15 minutes later and 5 points higher, at 390.83. Paying those 5 points is not a tax — it is the entire difference between a read and a guess, which is the argument made in full in the entry confirmation checklist.
The uncomfortable part: the index did the same thing
Here is what the Judas story leaves out, and it changes the conclusion.
At the exact same moment, the Nasdaq did it too:
| TSLA | QQQ | SPY | |
|---|---|---|---|
| Open | 392.35 | 712.01 | 752.74 |
| Low | 385.32 | 707.00 | 750.67 |
| Drawdown | -1.79% | -0.70% | -0.28% |
| Swept prior-day low | Yes | Yes (710.27) | No (750.25 held) |
| Reclaimed + reversed | Yes | Yes | Yes |
QQQ swept its own prior-day low and reclaimed it on the same clock. TSLA's drawdown was 2.6 times QQQ's — which is roughly TSLA's beta to the Nasdaq, not evidence of anything unique.
So the honest description of 16 July is not "smart money hunted TSLA's stops." It is: the market did an opening sweep-and-reverse, and TSLA expressed it with 2.6x amplification. The stock-specific raid narrative is doing no work here that beta was not already doing.
This has three consequences most Judas material never reaches:
- Your observations are not independent. If you count TSLA Judas days to build a statistic, you are mostly counting market opening-reversal days. Twenty observations clustered on ten market events is not twenty samples — and every sample-size rule below assumes independence you do not have.
- The instrument is a risk choice, not an edge. If the index is doing it too, the same thesis is expressible in QQQ at 40% of the volatility. Choosing TSLA is choosing leverage.
- Check the index before you call it a raid. A single name reversing against a falling market is a real, rare, idiosyncratic event. A name reversing with the market is Tuesday.
Grading the setup
Once the classifier says Judas, quality still varies. Weight the criteria — and weight them unequally, because evidence the market produced is worth more than a level you drew.
Load the TSLA preset and it scores 14 out of 14. Every box: both levels swept, volume climax, one-bar reclaim, displacement, structure shift, pullback held, index supportive, no bearish catalyst on the tape. A perfect card. And it paid — 385.32 to 395.31 inside an hour.
Which brings us to the only part of this article that actually matters.
The win rate you do not have
That 14/14 setup worked. This tells you almost nothing about the next one, and believing otherwise is how the whole genre goes wrong.
Search for Judas swing statistics and you will find confident numbers — "70% of opening sweeps reverse" — with no definition, no sample, no interval. Those numbers are decoration. There is no reliable published win rate for this pattern, for a reason that is structural rather than lazy: "Judas swing" has no standard definition, and neither does "reverted to value." Value can mean session open, anchored VWAP, premarket midpoint, opening-range midpoint, or prior-day POC. Different choices produce different win rates from the same tape. A number without both definitions pinned down is not a statistic.
So what should you believe about a single perfect example? Run the arithmetic on the inference itself:
One observation is consistent with a true rate of 21% — a catastrophic strategy — and also with 100%. Widen the sample and it barely improves:
| Your sample | Observed | 95% CI | Verdict |
|---|---|---|---|
| 1 of 1 | 100% | 21% – 100% | An anecdote |
| 3 of 5 | 60% | 23% – 88% | Still an anecdote |
| 9 of 15 | 60% | 36% – 80% | Includes "no edge" and "great edge" |
At n=15 — more Judas setups than most traders see in a quarter — the interval still spans from coin flip to excellent. You cannot tell those apart. That is not pessimism, it is the arithmetic of small samples.
How many do you need? To distinguish a true 60% from a coin flip with 80% power at the 5% level:
| If the true rate is | Setups needed |
|---|---|
| 55% | 783 |
| 60% | 194 |
| 65% | 85 |
| 70% | 47 |
At roughly one qualifying setup a week, n=194 is about four years of waiting — and that is before the correlation problem above eats into your effective sample. This is the real argument for pooling across many tickers and automating the classification: not elegance, but the only route to a sample size that can answer the question inside a career.
What exactly should I log to test this myself?
Freeze the definitions before you collect anything, or you will fit them to the outcome afterwards. Pin down: the sweep universe (which levels count), the excursion threshold (for example, at least 0.75x the 20-period 5-minute ATR beyond the level), the volume condition, the reclaim window in minutes, and the single target definition for "value" — pick one, do not test six and report the best. Then log, per setup: time to reclaim, time to target, maximum adverse excursion before the target, maximum favourable excursion, and the outcome. The two fields people skip are MAE and time-to-target, and they are the two that decide whether the trade is survivable with a real stop and a real option expiry.
Why is future leakage such a problem for this specific test?
Because the most tempting definition of value — the volume-profile point of control — is only knowable at the end of the day. If you enter at 09:45 and measure success against the session's final POC, you are grading with information that did not exist at entry, and your backtest will look far better than the strategy can ever trade. Every input has to be computed from data available at the decision moment: POC as of 09:45, VWAP as of 09:45, and nothing later. This one error probably accounts for more "amazing" opening-reversal backtests than every other mistake combined.
Why the option did not pay like the chart did
The underlying move and the options P&L are different trades, and the gap between them is where this setup quietly loses money.
Take the natural expression of that morning: a 390 call bought with the stock near 386, during the sweep. Bought at the option's own low it printed 2.90 and peaked at 7.95 — up 174%, a far bigger number than the stock's 2.6%. That leverage is the attraction, and it cuts in both directions.
TSLA 16 Jul 2026 — the full session, 30-minute bars (real SIP data)
The stock closed at 391.17 — still above the strike. The call finished in the money. And it still gave back close to half of what it made:
These are the contract's real traded prices (TSLA 17 Jul 390 call, 1DTE), not a model:
The trade was right, stayed right, and still surrendered 43% of its peak — because the premium topped when the structure topped at 10:45, not when the session ended. After 11:00 the market distributed, broke down to 386.86 by 13:30, and balanced. Those hours bled against a position whose thesis had already played out.
Why did the call only lose 43% when its intrinsic value lost 78%?
Because intrinsic is not the option. At the 10:45 peak the stock was 395.31, so the 390 call held $5.31 of intrinsic and $2.64 of extrinsic. At the close the stock was 391.17 — intrinsic collapsed to $1.17, a 78% fall. But the call did not fall 78%, it fell to $4.53, because extrinsic value is largest when an option is at the money. As price drifted back toward 390, extrinsic rose to $3.36 and cushioned the intrinsic collapse. Modelling this trade on intrinsic alone would have overstated the damage by roughly two to one — which is exactly why the numbers above are traded prices rather than arithmetic.
The Judas thesis expires when the markup expires. The structure said so at 11:00: failure to make new highs, then a break of the rising structure. That was the exit, hours before the close and 4 points better.
Do not buy the low. You cannot know it is the low. Buy the proof, pay the points it costs, and exit when the structure that justified the trade breaks — not at the closing bell.
And hold the win rate loosely: one perfect example is a 95% interval of 21% to 100%. Log your own, define "value" before you count, and let the data move the line.
The one thing to do differently tomorrow: when the opening drive sweeps a level, start a timer instead of an order. Write down the level, watch for a close back above it, and note how many minutes it took. Inside 15 with displacement, you have a setup worth an entry on the pullback. Past 20 with no reclaim, you have a trend day and you were about to fade it.
That timer is the whole edge — or, more honestly, it is the only part of this you can verify is real before you have four years of data.