Combining tools: a confluence workflow

It is 10:40 ET (22:40 ICT, winter). NQ opened with a drive higher and has pulled back to a shelf it built at 18,031–18,032. You have seven tools open. The footprint looks like buyers are defending, delta at the shelf is negative, the DOM shows a bid that keeps refilling, the tape is quiet, and price has not moved in twenty minutes. Three tools say buy, two say sell, two say nothing.

More tools did not give you more clarity. They gave you more votes, and votes are not analysis. This lesson is the workflow that turns a toolbox into a read: a fixed order, a witness list you write before you look, a drill that shows you where your tools disagree, and an honest account of what stacking witnesses does and does not do.

Key terms

Confluence (hội tụ): two or more separate references or signals pointing at the same price or the same conclusion Definition. The word describes agreement; on its own it says nothing about whether agreement improves the outcome.

Witness (nhân chứng): one lens on the auction (the value map, the auction structure, the footprint, the tape) giving its own testimony about what happened at a price Definition. Two witnesses are independent (độc lập) only when they are built from different inputs.

Chart first: four witnesses, three endings

How it works

The order of the read

Always read in this order House rule:

1. Value map (M2): where are we? POC, value area, HVN/LVN, composite references.
2. Auction state (M1): what regime? Balance or imbalance, who has been in control, who is trapped?
3. Tools (M3): what are the fingerprints saying? Footprint, delta, DOM, tape.
4. Price: what is the verdict? Acceptance or rejection?

Tools come third because a fingerprint needs a location and a regime before it describes anything: a burst of selling means one thing at the low of a balance and another in the middle of a trend leg. That is M3.01’s point, used here as a working rule, not a law. Price comes last because it is the only witness that settles anything. When the footprint shows heavy buying but price cannot rise, that is effort without result (M3.02): the buyers are not getting paid. Believe the price, not the description.

This order is a process choice. It keeps you from reading a footprint in a vacuum and from trading a story that price has already rejected. It is not a tested edge, and the sections below say why that matters.

Independent witnesses, not repeated ones

A read worth acting on has testimony from several independent lenses, each built from different inputs House rule. Delta, CVD and the delta profile are one witness: all three are the same aggressor volume, summed three ways. A footprint showing selling into a bid and a DOM showing that bid refilling are also one witness at that moment: both describe the same resting order. A useful list draws on different inputs:

• Level from the map (prior VAH, today’s shelf);
• Structure from the profile or TPO (a poor high, a balance edge);
• Order flow from the footprint or DOM (aggression meeting a passive order);
• Tape (bursts with or without follow-through).

Decide that list before you look at the chart, take every testimony, and write down the ones that disagree.

Contradictions are information

When the footprint shows aggressive buying and price falls, you have not found a broken tool. You have found buyers being absorbed (M3.02), which tells you more than agreement would have. The witness list does not exist to make every tool agree. It exists to make every tool testify, so that price can deliver the verdict and you can see in advance which verdict would kill the read.

What stacking does not do

The popular claim is that each extra witness raises the odds: two references at one price make a “strong level”, and a full house of level, divergence and absorption is a high-quality trade. It has been tested twice, and it failed both times.

Levels first. Our lab scored pre-market levels against distance-matched random prices over 721 sessions. Zones where two or more level families sat together were respected no more often than single levels: the difference ran from −0.1 to +0.7 pp, with intervals of about ±4 pp Measured: no edgeNasdaq index. In the same study no single level family beat random prices either.

Order flow second. One trader’s public NQ study tested a full-house confluence of level, delta divergence and absorption or exhaustion. It beat its baseline by 3.8 pp in development (interval −2.8 to +10.5, n = 414) and lost 6.8 pp in the holdout LoreOne trader's study. A search over every pair and triple of order-flow elements found no more 3-point effects than the same search on shuffled outcomes.

Why the workflow is still worth running

If agreement does not raise the odds, why keep the list? Because it does three other jobs. It tells you where and when to pay attention, so you are not reacting to every print. It forces you to name the price at which the read is wrong, which is where the stop goes. And it stops you from rationalising: a list written before you look cannot be bent toward the trade you already wanted. Footprint signals of this kind also tend to flag that a bigger move is coming without saying which way: in the same public study they widened both the best and the worst excursion PlausibleOne trader's study. A stack tells you that something is about to be decided. It does not tell you the decision.

Worked example: NQ, the full stack (synthetic session)

Synthetic session; the levels are illustrative.

• Map (M2): price 18,032.00, at this morning’s shelf, inside today’s value (POC 18,025.00, VAH 18,041.00), above the composite HVN.
• Auction (M1): the morning was an open drive up; since 10:15 ET (22:15 ICT) price has balanced under VAH. Buyers have been in control; the open question is acceptance above 18,041.00.
• Order flow (M3): from 10:38 ET about 1,250 contracts sell into the bid at 18,031.00–18,032.00 and price gets no lower than 18,030.75; the DOM shows the bid refilling. One witness: effort without result at the shelf.
• Tape: after the selling, prints at the offer resume and no seller follows through.
• Price: holding above the shelf for twenty minutes.

Four lenses, one testimony: responsive buyers are defending the shelf. The plan is conditional. If price trades back through 18,034.00, go long 1 NQ with the stop at 18,027.75, just under the shelf: 25 ticks, $125 of risk before costs. The first target is VAH 18,041.00 (28 ticks, $140), which is the real test. If price accepts below 18,028.00, the read is dead: the stack dissolves and the shelf-flip logic of M2.06 takes over. Every read carries its own death condition.

The mirror and the failure

Mirror. The same list read at VAH 18,041.00 on a day that has balanced under it: the offer refills while buyers lift into it without progress, the tape shows buy bursts with no follow-through, and the structure shows a poor high. If price trades back below 18,038.00, the short has its stop above the shelf and its first target at the POC. Same workflow, other side.

Failure. Another synthetic day, the same four witnesses at the same shelf. At 10:52 ET (22:52 ICT) the bid at 18,031.00 stops refilling. Sellers trade through 18,028.00 in two bars and the long is stopped for 25 ticks plus costs. Nothing was misread: the witnesses described the place and the moment correctly. They could not tell you whether the resting buyer would stay, and that is exactly what the holdout result above says. The plan still did its job, because the loss was the size you chose before entry.

The one-chart drill

Drill: three passes, then merge

Pick one instrument and one session. Narrate it three times, one lens per pass, and write each narrative down before starting the next.

1. Profile only (M2): where was value, what was the structure, where did it change?
2. Footprint only (M3.01–M3.07): where was the aggression, and where did it fail to move price?
3. DOM and tape only (M3.08–M3.10): what did liquidity do, where were the bursts, and did they follow through?
4. Merge the three narratives and list every place where two passes disagree.

Common mistake: Confirmation shopping. You run through tools until three agree with the trade you already wanted, then stop. That is a rationalisation stack, not a witness list. Decide the list before you look (level, structure, order flow, tape), take every testimony, and let the disagreements veto the trade.

Common mistake: Counting one lens several times. Delta, CVD and the delta profile agreeing is one witness, not three; a footprint and a DOM describing the same resting order is one witness too. If your stack is mostly one input, it is a single opinion repeated.

Common mistake: Treating agreement as a probability (an evidence misread). “Four witnesses agree, so this works most of the time” is the claim the data above rejected. Many taught confluences also fire only a handful of times a year, so screen time alone can never tell you whether they work Definition. Size the trade as if the stack adds nothing, because on the evidence so far it does not.

What we don’t know yet

• Whether a fixed, pre-registered witness list on NQ futures beats a placebo when each trade is scored by its own invalidation and target: not yet measured.
• Whether pairs that include a level touch carry a small effect: in the practitioner search they sat slightly above the chance mean (58 vs 46.6), which deserves a session-clustered retest; not yet done.
• Order-flow confluence on the lab’s own NQ intraday data: not yet measured; it needs the NQ tick export the lab is still building.

Recap:

• Read in a fixed order: value map, auction state, tools, price verdict. It is a working rule for discipline, not a tested edge.
• Count independent lenses only; tools built from the same input are one witness. Contradictions are information, and price settles them.
• Stacking adds nothing measured: confluent levels matched single levels over 721 sessions, and full-house order-flow confluence lost its edge in the holdout on NQ.
• Every read carries its invalidation; size it as if the stack adds no probability.
• The one-chart drill (three single-lens passes, then merge) is how you find where your tools disagree.

Further reading:

• Order Flow, Measured: all claims: one trader’s public NQ study; the confluence rows DN-04, DN-12 and DN-13.
• Our lab: pre-market levels and confluence zones were touched, respected and stalled at the same rates as distance-matched random prices over 721 sessions.
• How To Add Depth To Your Trading Tools: finished and unfinished auctions; observation-first reasoning.
• Is Order Flow Trading Overrated?: information versus technique, and why one-rule systems fail.
• Momentum vs Mean Reversion: Which Dominates in a Choppy Market?: the regime question behind the second level of the read.

This lesson connects to: M4.08 (morning sheet, where the witness list is written before the open) · M4.09 (order-flow confirmation written so it can be scored) · M5.13 (telling setups apart) · M3.16 (why the list needs few tools, not many). M3.16 comes next because a witness list only works when you can read every lens on it.

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