Setup vs entry: the four entry families

It is 10:05 ET (22:05 ICT, winter). NQ falls into yesterday’s value area low at 18,040, the level you marked before the open. You buy the first touch on impulse and pick a stop 10 points lower after the fill. Price spikes 12 points under the level, takes your stop, closes the next bar back above the level and then rallies 35 points without you. Was the read wrong, the entry wrong, or neither? This lesson separates the two decisions so that each can be written down and judged on its own.

Key term

A setup (mẫu hình) answers where and why: the context, the level, the auction story Definition. An entry model (mô hình vào lệnh) answers when, where you are wrong, and where you get paid: a trigger, an invalidation and a target written as one unit before the fill Definition. A trigger with no pre-set stop is not a model; it is an impulse.

The four entry families

Every entry you take belongs to one of four families. What separates them is arithmetic, not prediction: the earlier you commit, the closer your price is to the invalidation; the later you commit, the more of the auction you have seen Definition.

1. Breakout (phá vỡ): enter when price trades through a level or a recent high or low. You see the move start, and you pay a worse price for it (M6.02).
2. Pullback / retest (hồi lại / kiểm định lại): enter on the return to a level that just broke. A better price than the breakout, but moves that never look back leave you out (M6.03).
3. Rejection (từ chối): enter at or just after a failed probe at an extreme. The best price, but you trade against the move that is happening, and a resting limit sets its stop before the probe has finished (M6.04).
4. Continuation (tiếp diễn): join a move already under way, on a pause or flag. It needs a move to exist first (M6.05).

There is no free family. “Confirmation” here means only that you have seen more before committing; it is not a measured improvement in your odds. Whether seeing more raises the win rate enough to pay for the worse price is the open question of this module (next section).

The setup does not make the entry safe either. Our lab tested pre-market levels as a group, prior value-area edges and overnight extremes included, against random prices at the same distance: they were respected no more often Measured: no edgeNasdaq index. The level is where you choose to look, not a reason price turns. So “I read it right but entered wrong” is a hindsight claim: one trade cannot tell you whether the read, the entry or plain noise failed. Only a logged sample of the same model can.

What the data says

So treat the four families as four ways to define risk, not four edges. The lesson’s core content (what a setup is, what an entry model is, how price trades against seeing more) is definition and arithmetic Definition.

Worked example: three entries on one setup

Synthetic NQ day, the figure above. Yesterday’s value area low is 18,040.00. All three models use the same target, 18,075.50, written before the open. 1 NQ = $20 a point; 1 MNQ = $2.

• A, rejection with a resting limit. Buy limit 18,040.00, stop 18,030.00, a fixed 10 points because the probe’s low is not known yet. Risk 10 points = $200 on 1 NQ. Reward 35.5 points = 3.55R. Break-even win rate: 10 ÷ 45.5 = 22% Definition. On this path the probe reaches 18,027.75 and A is stopped out.
• B, rejection with a close trigger. Trigger: a 5-minute close back above 18,040 House rule. The 10:10 bar closes at 18,045.25; buy 18,045.50 at the next open. The probe low is now known, so the stop goes one tick under it: 18,027.50. Risk 18 points = $360. Reward 30 points = 1.67R. Break-even: 18 ÷ 48 = 37.5%.
• C, breakout of the rejection bar. Buy stop 18,050.00, one tick above the 10:10 bar’s high of 18,049.75; same stop 18,027.50. Risk 22.5 points = $450. Reward 25.5 points = 1.13R. Break-even: 22.5 ÷ 48 = 46.9%.

The setup never changed. Each later entry saw more of the auction and paid for it: C needs to win more than twice as often as A just to break even, before costs (M6.08 adds them). The data above does not say any of the three clears its bar.

A is still an entry model. It has a trigger (the limit price), a stop and a target written in advance, so it can be scored. The trader in the opening scene had none of those; that, not the first touch, was the mistake.

The mirror and the failure

Mirror. Flip the figure: price probes above yesterday’s value area high and returns inside. A sells a limit at the level with a fixed stop above; B sells after a 5-minute close back below the level, stop one tick above the probe high; C sells a stop one tick under the rejection bar’s low. Same arithmetic, same trade-off.

Failure. On another day the probe below 18,040 keeps going and is accepted: price holds under the level and the next bars close lower. A loses its 10 points. B and C never trigger, so they lose nothing. On a third day price touches 18,040 and rallies at once without a close below: A wins its 3.55R, B never fires, and C fires only if price takes out its trigger bar’s high. Each model trades a different set of days. That is why you compare models on a logged sample, never on the one chart where your favourite worked.

One model at a time

An entry model is only complete with its stop and target logic (M6.08): trigger, invalidation and objective are one unit. Learn the families in order, trade one on the sim until its failure modes are familiar, then add the next House rule. Collecting entries you cannot execute is worse than knowing one well. How limit and stop orders actually fill, and the prop-firm rule on simulated fills, are in M6.06.

Common mistake: Entering on the idea of the setup (“it is rejecting, buy”) with no trigger, no stop and no target. The setup told you where; without a model you have no when, so every tick feels like the entry and every adverse tick feels like betrayal.

Common mistake: Treating the chart where the confirmed entry won as proof that confirmation works. It is one path. The same rule skips some winners as well as some losers, and on NQ no family has a shown edge yet.

Common mistake: Moving the stop after the fill. If the invalidation was not written before the click, the entry had no model, whatever the trigger was.

Drill: three models, ten sessions

Pick one level you mark every morning (yesterday’s value area low or high). Replay ten sessions in ATAS X.

1. Before each session, write models A, B and C for that level exactly as in the worked example, with one shared target.
2. At each test of the level, log for each model: triggered or not, entry, stop, exit, result in R.
3. Compute each model’s win rate and average R, and compare each win rate with its break-even.

Recap

• A setup is the read (where and why); an entry model is trigger, stop and target written as one unit before the fill.
• The four families (breakout, pullback/retest, rejection, continuation) trade price against how much of the auction you see first; that trade-off is arithmetic, not an edge.
• On the same setup, later entries carry more risk and a higher break-even win rate.
• No entry family, and no confirmation at a level, has a shown edge on NQ yet; judge a model on a logged sample, not on one chart.

Further reading

• Algorithmic Trading: Winning Strategies and Their Rationale (Ernest Chan): why every strategy needs an explicit rationale before it earns capital.
• Tharp Think: Trading Concepts: the argument that the system around an entry matters more than the entry itself.
• Order Flow, Measured: All Claims: one practitioner’s public, pre-registered tests of NQ order-flow claims.
• Our lab: pre-market levels as a group were respected no more often than random prices at the same distance (721 sessions, Nasdaq-index proxy); order-flow confirmation at a level is not yet measured.

Connects to: M5 setups (your entries execute them) → M6.02, the first family: breakouts.

Educational content only — not financial advice. Futures trading involves substantial risk of loss.