Telling setups apart

It is 11:05 ET (23:05 ICT, winter). The ES footprint prints a clean delta divergence at 6,025.00: a higher high on visibly weaker buying than the push 15 minutes earlier. It is the shape from M5.04, and you short it. Price runs 15 points through you. The pattern matched the lesson. What you skipped was the step before the pattern: asking whether this moment had the context that a reversal setup relies on. This lesson is that step, a short written classifier, and an honest account of what it can and cannot do for you.

Key term

Setup classification (phân loại setup): before any trade, name the mechanism the setup relies on and check, with written questions, whether the context that mechanism needs is present now House rule. Reversal setups (M5.01–M5.06) rely on responsive traders defending an extreme. Continuation setups (M5.07–M5.12) rely on initiative traders being accepted beyond value. A setup whose context is missing, or unclear, is unclassified (chưa phân loại) and is not traded.

The classifier: one mechanism, three questions

Start with the mechanism, because it tells you what context to look for. A reversal setup is a claim that responsive traders will defend an extreme. A continuation setup is a claim that initiative traders have been accepted beyond value and will keep going. Each setup lesson in this module wrote its context requirements into the rule. The classifier checks them in the same order every time House rule:

1. Regime (chế độ thị trường): what is your written day-type hypothesis right now (M4.03)? A trend hypothesis is the context continuation setups ask for; a balance or neutral hypothesis is the context reversal setups ask for. Treat it as a hypothesis with a kill condition, not a fact: the label arrives at the close Definition. In 58 recent NQ sessions our lab found the label so far at 11:00 ET matched the closing label on 24 Context, and a public practitioner study found no skill in forecasting the day type by 11:00 LoreOne trader's study (M4.03 has both numbers).
2. Location (vị trí): is price at a level you mapped before the open, or mid-range, inside value, at a VWAP pullback? Reversal setups need the first; continuation setups usually happen in the second. The map tells you where to watch, not where price turns: mapped levels and confluence zones were respected no more often than random prices at the same distance Measured: no edgeNasdaq index (the box in M5.01).
3. Acceptance (chấp nhận): has a level been probed and then rejected back inside, or accepted beyond, by the checks you wrote in M4.01 Definition? This is the only question that reads the live auction. A probe that is still deciding has no verdict yet.

Three answers that all fit one mechanism make a candidate for that setup. Then the setup’s own trigger, stop and target still have to fire. The working rule: when the pattern and the context disagree, you skip House rule. It is a rule about what you are allowed to trade, not a finding that context makes trades win. None of the three answers has a measured edge on its own, and the combination has not been tested as a filter either. What the rule buys you is consistency: every trade in your log was taken under the conditions its rule was written for, so the log can be scored.

The same pattern, opposite labels

• A delta divergence at a mapped extreme, on a neutral-day hypothesis, after a probe was rejected: reversal candidate for M5.04. The divergence itself showed no edge on NQ LoreOne trader's study; the classifier only says the context the setup needs is there.
• The identical divergence mid-range on a trend-up hypothesis, with no level tested: not a reversal setup. Shorting it means fading the initiative your own hypothesis describes.
• A stack of buying imbalances breaking a balance edge, with the break accepted: continuation candidate for M5.09, whose stacked-imbalance trigger also showed no edge on NQ LoreOne trader's study.
• The identical stack stalling into a mapped high on a neutral hypothesis, with no acceptance above: unclassified until the probe is accepted or rejected.

The footprint draws the same picture in each case. The classifier decides which written rule, if any, the picture belongs to.

Myth / hypothesis: “Context wins: the right pattern in the right context is a high-odds trade.” Lore Context answers which rule applies. It does not supply odds. The levels that define location tested at the placebo rate, the day type is only known at the close, and stacking context elements did no better than chance in the study below. Whether classifying first raises the win rate of any M5 setup on NQ is not yet measured.

Why it is still worth knowing

• It stops rule-shopping. Without a classifier, every pattern can be traded as whichever setup fits the trade you already want.
• It frames risk. A reversal candidate has a nearby, objective invalidation at the extreme; a mid-range “reversal” has none.
• It makes your journal testable. Every logged trade carries its three answers, so you can later compare classified trades against the ones you skipped.

Transition zones

The hardest moments are the transitions: a failed breakout that turns into acceptance (M5.01 becoming M5.07), or a pullback that slices through VWAP instead of holding it (M5.08 becoming a reversal question). Classify after the acceptance check, never during the wobble. While a probe is still deciding, question 3 has no answer, so the setup is unclassified and you wait. If any of the three answers is “unclear”, the setup is unclassified and you skip it House rule. A skipped trade costs nothing except the trade you might have had, and you can measure that cost later from your log.

What the data says

The popular promise behind confluence trading is that the more context elements agree, the better the trade, and that experience teaches you which combinations work. Two problems: the combinations tested no better than chance, and they are too rare to learn from by watching.

So the classifier is graded as what it is: a house rule that keeps your trading consistent and scorable. The edge of any setup it lets through is still that setup’s own, unproven, grade.

Worked example: the divergence that was not a short

Synthetic session, ES (0.25-point ticks, $12.50 a tick, $50 a point); it shows the reading, not a signal.

• Map, written before the open. Prior-day high 6,041.00; overnight high 6,018.00; nothing else within 15 points of 6,025.
• 09:30–10:30 ET. Open drive up from 5,990.00; initial balance 5,990.00–6,012.00. VWAP never threatened.
• 10:40 ET. The IB high breaks and the break is accepted on the M4.01 checks. Written hypothesis at 11:00 ET: “Trend up. Kill: a 30-minute close back below 6,012.00.”
• 10:50 ET. Push to 6,023.50 on strong buying.
• 11:05 ET (23:05 ICT, winter). New high at 6,025.00 on weaker buying and light volume: the divergence.

Run the classifier. Mechanism: a divergence short is a reversal setup, so it needs a responsive extreme. (1) Regime: the hypothesis is trend up, the context continuation setups need ✗. (2) Location: 6,025.00 is 16 points under the nearest mapped level ✗. (3) Acceptance: no level was probed, so there is no verdict ✗. Verdict: not a reversal candidate. You skip the short. The classifier does not hand you a long either; a long needs a continuation setup’s own trigger, such as a breakout–retest (M5.07).

Outcome on this synthetic day. By 11:30 ET price is accepted above 6,028.00 and trades 6,040.00 at 12:10. The skipped short would have lost. One day is one outcome, not evidence. In the journal it goes under “skipped by rule”, with the three answers written next to it.

The mirror and the failure

The long side (mirror). Flip everything. A bullish divergence (a lower low on weaker selling) at 11:05 ET, mid-range, on a “trend down” hypothesis with no level tested, is not a reversal long: skip it. The same bullish divergence at a mapped low on a neutral hypothesis, after a probe below the low was rejected back inside, is a reversal candidate long, with the stop beyond the probe low.

The failure. Another synthetic morning, the same three answers, the same skip. This time the trend-up hypothesis is wrong: price stalls at 6,025.00, the 11:30 ET period closes back below 6,012.00, the kill condition fires, and price falls to 6,005.00 by 12:15. The skipped short would have made 20 points. The classifier did its job, which was to stop a trade whose context was missing at 11:05; it did not forecast the turn, and it cannot. When the kill condition fires, reclassify on the new information: the day is now a neutral or failed-trend hypothesis, and the next reversal candidate needs a mapped extreme and a verdict, not the old divergence. Count these misses in your log; they are the price of the filter.

Common mistake: Pattern-matching without classifying. Each setup lesson teaches a pattern together with its context requirements, but under pressure the brain keeps the pattern and drops the conditions. Ask the three questions aloud, or write them, before every setup. It feels slow at first and gets quicker with repetition.

Common mistake: Treating “I have seen this work many times” as a base rate. A confluence that fires six times a year gives you a few dozen examples over a trading career, and memory keeps the dramatic ones. That is a story, not a measured rate. Write the rule and score it.

Common mistake: Reading the classifier as an edge. “Right pattern, right context” sounds like high odds. On NQ, the context elements and their combinations tested at chance. The classifier tells you which rule to apply; the rule’s own evidence box tells you how much to expect from it.

Drill: classify thirty setups, then count your favourite confluence

1. Before the session, write your map and your day-type hypothesis with a kill condition.
2. In replay, stop at 30 moments where any M5 pattern prints. For each, write the mechanism and the three answers before you look ahead.
3. Label each one: reversal candidate, continuation candidate, or unclassified. Then record what price did over the next hour.
4. Pick the confluence you trust most. Count how many times it fired in the last 12 months of data you can replay.

What we don’t know yet

• Whether classifying first raises the win rate of any M5 setup on NQ, compared with taking every occurrence of the same trigger: not yet measured.
• Whether a day-type hypothesis written at 11:00 ET changes the outcome of reversal or continuation setups: not yet measured; forecasting the day type by 11:00 showed no skill in the public study.
• One related lead is still being tested: a delta divergence into a level’s first touch (not the second test) was graded Lore with a promising holdout in the public study. It needs a forward test before it counts.

Recap

• Classify by mechanism: reversal setups need a responsive extreme, continuation setups need initiative accepted beyond value.
• Check the context with three written questions: regime hypothesis, location on the map, acceptance verdict. When pattern and context disagree, skip.
• Transition zones and unclear answers are unclassified: wait for the verdict, or do not trade.
• The classifier is a house rule for consistency, not an edge: levels, day-type forecasts and stacked confluences tested at chance on NQ and the index proxies.
• A confluence that fires 4–10 times a year cannot be validated by screen time; write it as a rule and score it.

Quiz

Further reading

• Order Flow, Measured: all claims: the public practitioner study behind both evidence boxes (DN-04, DN-12, DN-13).
• How To Add Depth To Your Trading Tools: letting the market’s behaviour, not the pattern, decide which reading applies.
• Momentum vs Mean Reversion: Which Dominates in a Choppy Market?: regime as the first question.
• Is Order Flow Trading Overrated?: patterns without context as a failure mode.
• Our lab: 226 pre-registered tests on 721 sessions found no pre-market level family, and no confluence of families, respected more often than random prices at the same distance.

Prerequisites: M4.01 Acceptance, M4.03 Day types in practice, M5.01 Failed breakout at a level, and EV3.01 on sample size. Next: M5.14 Entry patterns with R:R attached, which turns a classified candidate into an entry with a stop, a target and the break-even arithmetic.

Educational content only, not financial advice. Futures trading involves substantial risk of loss. Synthetic examples are illustrations, not trade calls.