What Makes a Trade Setup High Probability? The ICT/SMC Criteria

Six things are supposed to make a setup high probability, and most traders who use the ICT and smart money vocabulary can name all six from memory. Higher timeframe agreement, market structure, an order block, a fair value gap, a liquidity sweep, and the right session. Naming them is the easy part. The hard part is that a chart with four of them and a chart with one of them feel roughly the same at 9:41 in the morning, because the feeling comes from wanting the trade rather than from the count. So the version of this question worth answering is arithmetic. How many of the six are actually on the screen, how many of those are independent of each other, and does the answer change anything about how you take the trade.

Quick Answer

What Does High Probability Actually Mean?

The word probability implies a number, and nobody trading a discretionary setup has that number for the chart in front of them. What the phrase describes in practice is setup quality: how much of the chart agrees with the direction you want to trade before you click. Each condition on its own is close to a coin flip. An order block by itself is a rectangle you drew. A sweep by itself happens forty times a session inside a range. Agreement between conditions that came from different places is the thing worth paying attention to, and agreement is countable.

Countable is the useful property here, because it survives being tired, tilted, or three hours into a session with nothing to show for it. A criterion is either on the chart or it is not, and a fixed scoring order stops you from grading the setup after you have already decided. That is the entire argument for scoring a trade before you enter it rather than reconstructing your reasoning afterwards, when the outcome has already coloured it. If you are still choosing which setups to trade at all rather than how to rate them, the catalogue of day trading strategies sits upstream of this page.

Everything below is asset agnostic on purpose. These six criteria are the generic ICT and SMC checklist, and they read the same on an index future, a currency pair, or a large-cap stock. The instrument-specific version, where gold's $50 round-number spacing, tick volume instead of real volume, and news sensitivity all change what a good chart looks like, is in the walkthrough on grading a XAUUSD setup. Read this one for the framework and that one for how the framework bends around a single instrument.

What Are the Six ICT/SMC Criteria?

Here they are with the boundary that matters most: what actually counts as met, versus the generous version of the same criterion that everyone awards themselves at 9:41. The third column is where each one tends to break, because knowing the failure mode is what stops you scoring it wrong twice.

A few of those need the actual definition attached, so here is the short version of each. Higher timeframe alignment uses a multiple of roughly four to six: enter on the 5 minute and the 1 hour is the chart that has to agree, enter on the 15 minute and it is the 4 hour. The step-by-step version of that top-down read, including what to do when the two charts point opposite ways, is in the piece on reading a chart from the higher timeframe down. Market structure is the series of swing points: higher highs with higher lows is an uptrend, lower highs with lower lows is a downtrend. A break of structure is price taking out the previous swing point in the direction of the trend, so it continues the story. A change of character is the first break against the trend, and it is a warning that the story is changing rather than a cheaper entry into the old one.

The middle three are the ones the ICT vocabulary is actually built around. An order block is the last opposing candle before the impulsive move that broke structure: the last red candle before the rally that took out the high, or the last green candle before the drop that took out the low. A fair value gap is a three-candle imbalance where the first candle's wick and the third candle's wick do not overlap, leaving a price range the middle candle ran straight through without trading properly. Liquidity is where stops rest: buy-side liquidity above equal or obvious swing highs, sell-side liquidity below equal or obvious swing lows, and a sweep is price spiking through one of those levels and snapping back inside on a wick. Full definitions with worked examples of each live in the breakdown of how smart money concepts are read off a chart, and if you want to see the whole vocabulary inside one educator's complete system rather than as a scoring rubric, the write-up on what TJR teaches covers one in detail.

A concrete version, using round numbers so the geometry is easy to hold. Price grinds down and puts in two lows at 19,850, close enough together to call them equal. The next push down wicks to 19,832 and closes back at 19,858, which is the sweep. The last red candle before the rally that follows spans 19,845 to 19,860, which is the bullish order block. That rally runs up through the prior swing high at 19,910, which is the break of structure. Inside it, one candle's high sits at 19,878 while the candle two later has its low at 19,894, so the range between them never traded properly and that is the fair value gap. Price then pulls back into 19,894 to 19,878, and that retrace is the entry zone. Five criteria, one sequence.

Session is the one traders coming from a purely technical background tend to skip, and it is a participation criterion rather than a chart one. Structure needs someone on the other side of it to resolve, and a textbook order block at a dead hour is a pattern with nobody trading against it. US equities have one obvious answer, since the NYSE regular session runs 9:30 to 16:00 ET and the flow concentrates at the open and into the close. Currencies never close, which makes the criterion feel meaningless until you look at when the volume is actually there. The BIS triennial survey measured $7.5 trillion of daily OTC foreign exchange turnover in April 2022, and that total is spread very unevenly across the 24-hour cycle. The ICT kill zones are one way of naming the busy parts: London open around 02:00 to 05:00 ET, New York morning around 07:00 to 10:00 ET, and the overlap between the two where both centres are live at once.

Which Criteria Are Gates, and Which Are Bonuses?

Treating all six as interchangeable points is where most scoring systems go wrong. Four criteria out of six sounds identical whichever four you got, and it is not. Two of them decide whether the setup exists at all, and the rest only adjust how good a version of it you are looking at.

Which means the honest way to state a score is two numbers rather than one. Both gates passed, plus however many of the remaining four are present. A setup described as "gates plus two" carries information. A setup described as "4 out of 6" might be a clean trend pullback into an order block, or it might be a sweep and a gap on a chart that has been going sideways since the open.

Six Things, or One Thing Four Times?

Go back to the worked example above and count where each criterion came from. The sweep was the wick that triggered the reversal. The order block was the candle immediately before the leg that followed. The break of structure was what that leg did. The fair value gap was the hole the same leg left behind. One impulsive move produced four criteria, and describing a single event in four vocabularies is not the same as having four pieces of evidence.

There is a quick test for this. Ask whether each criterion could still have been true if that leg had not happened. Higher timeframe trend: yes, it came off a different chart and was already there yesterday. Session: yes, it came off the clock. A level that has been respected three times this week: yes, it predates the move entirely. The order block, the gap, and the break of structure: no, all three are downstream of the same candles. That distinction between signals that share a cause and signals that do not is the substance of counting confluences honestly, and it applies to indicator stacks in exactly the same way that it applies here.

What Should the Score Actually Change?

A score that changes nothing is decoration. The only variable it should touch is size, and the direction of that adjustment matters more than the size of it.

Notice that the response column tops out at full planned size. You cannot derive a position size from a confluence count, because that would require somebody to have run these exact six criteria across a few hundred of your trades and measured what each additional point was worth. Nobody has done that, including the people selling the checklist. What the score can honestly do is cap: take size off at a 3, refuse the trade when a gate fails, and never let a 6 talk you above your normal risk. The arithmetic that turns a stop distance into a share count is unchanged by any of this and is worked through in the walkthrough on sizing from risk per trade.

Two conditions sit outside the six and override all of them. The first is reward-to-risk. A 6-of-6 chart with the stop 40 points away and the nearest sensible target 30 points away is a losing trade with excellent scenery, and most people who trade ICT setups use a 1:2 minimum for exactly this reason. The mapping between a ratio and the win rate it quietly demands is laid out in the piece on what your reward-to-risk actually requires. The second is the candle close. A sweep that has not closed back inside the level is still in progress, and an order block that price is currently slicing through is not a zone, it is a former zone. Waiting for the close costs a few points of entry and removes a whole category of trades where you were early to a level that never held.

When Does the Checklist Fail?

Every scoring framework has conditions where the scoring itself stops meaning anything, and this one has five worth knowing before you trust a number it produced.

• Ranges
  Inside a sideways range, every touch of the edge looks like a sweep and every reversal candle looks like a reaction from an order block. You can find four criteria on a 1 minute chart of a two-hour range without a single one of them being informative. If the higher timeframe is flat, the gate has already failed, which is the checklist doing its job.
• Thin instruments
  Fair value gaps on a low-volume small cap or an illiquid pair are usually the result of nobody trading between two prices, not of aggressive one-sided flow. The candles look identical. The meaning is not. Same problem with equal highs on a stock that trades 40,000 shares a day: there are not enough stops resting there to be worth running.
• Cropped time axes
  A large share of the screenshots traders actually work from have the clock cut off. If you cannot read the session, score it absent. Assuming the session because the candles look busy is how a 3 becomes a 4 without anything changing on the chart.
• Criteria hunting
  Once you have decided you want the trade, the chart will produce an order block for you. This is ordinary confirmation bias, and it is the reason the scoring order has to be fixed and the score has to be written down before the entry rather than after.
• Sample shrink
  Demanding 6 of 6 might give you three trades a month, which is not enough to learn anything from. Every extra criterion raises the quality of each trade and cuts the number of trades you get to measure, and you need the measurements more than you need the perfect entry.

The fourth one deserves a moment because it is the expensive one. Confirmation bias is the tendency to search for and favour information that supports what you already believe, and a six-item checklist is an unusually comfortable place for it to hide, because every point you award feels like analysis. The counter is procedural rather than psychological: score in the same order every time, write the score before the entry, and record which criteria you scored as absent. Absent criteria are the useful data. Anyone can list the reasons they took a trade.

There is also the flat honesty that all of this is discretionary. Two competent traders will draw the same order block at slightly different prices and disagree about whether a wick counts as a sweep. The checklist reduces the disagreement, it does not remove it. What it genuinely buys you is consistency with yourself over time, which is the thing that makes a record of your own trades worth reading, and that record is what eventually tells you whether any of this constitutes an actual edge rather than a well-organised habit.

Where Does AI Fit Into Scoring These?

Reading these six criteria off a chart is pattern recognition on a static image, which is something AI-powered analysis is genuinely suited to. SnapPChart reads them directly off a screenshot you upload. The market structure read covers the swing highs and lows, whether the most recent move was a break of structure or a change of character, and how the candles reacted at the key level. Alongside it sit four independent ICT reads: where resting liquidity is, whether a sweep took it, at what price and on which side; a fair value gap with its direction, its price range, and whether it has been filled; an order block with its direction and its price zone; and the kill zone, which is only named when the chart's time axis is legible enough to identify the session.

The part that matters for scoring is what happens when a concept is not there. Each of those four is filled only when it is genuinely visible on the chart, and left blank otherwise rather than invented to complete the set. A blank field is a criterion you should score as absent, which is precisely the discipline that criteria hunting erodes when you are doing the counting yourself. The mechanics of how a screenshot becomes a structured read are described under AI chart analysis, and the wider question of what this class of tool can and cannot see is covered in the overview of how AI tools analyse charts.

The limitation is specific and worth stating plainly. This is a read of a static picture, not of live order flow. It can tell you that price spiked through the equal lows at 19,832 and closed back inside the level, because that is in the image. It cannot tell you that institutional buying caused the snap-back, because depth of market, the tape, and the actual orders behind the move are not in the image and never will be. It describes what the price action shows and stops there. It also does not stream, scan, alert, or place anything. You upload one chart, which means the higher timeframe gate is still yours to check unless you upload the higher timeframe chart as a second read, and the score you end up with is still a judgement you made rather than a verdict handed to you.