Data & settings behind the numbers

It is 16:00 ET on Thursday 1 October 2026 (03:00 ICT Friday). Two traders on the same desk write tomorrow’s NQ levels from the session that just closed. One writes POC 30,820.00. The other writes POC 30,634.25, almost 186 points lower. Neither made a mistake: the first profile counts every trade since the Globex open at 18:00 ET the evening before, the second counts only 09:30 to 16:00 ET.

Every tool in this module is arithmetic on trades. Before an order-flow number means anything, you need to know which trades went in, how they were labelled and grouped, and which settings turned them into a picture. This lesson is that checklist.

Key terms

Session template (mẫu phiên giao dịch): the start and end times a platform uses to build a session’s profile, VWAP or cumulative delta Definition. “RTH” is 09:30–16:00 ET; “full session” or “Globex” is 18:00 ET the evening before to 17:00 ET.

Print grouping (gộp lệnh khớp): how a platform turns the trades it receives into “one trade” for size filters and bubbles Definition. One aggressive order that fills against several resting orders can arrive as several prints with the same timestamp; summed, they are one larger trade.

Percentile threshold (ngưỡng phân vị): a cut-off taken from your own measured distribution, such as “the 90th percentile of 5-minute |delta| between 12:00 and 14:00 ET”, instead of a fixed number copied from someone else Definition.

Term you need: aggressor side. Each trade has a buyer and a seller; the aggressor is the one who crossed the spread. Delta and footprints are built from it, and M0.07 shows how much delta moves when a platform guesses the side instead of recording it. The roll (switching to the next quarterly contract) is covered in M0.08 and MKT1.02.

Chart first: one session, two templates

How it works

Four data checks, in this order

1. Is the side of each trade recorded? Footprints, delta, CVD and the tape all need tick data: every trade with its price, size and aggressor side Definition. A file of 1-minute bars keeps four prices and one volume number, so the inside of the bar is gone and no setting brings it back. A platform that draws a footprint from bars, or a replay that runs on candles, generates the inside of each bar; its imbalances are computed from generated numbers. Volume profiles are the exception: they need volume at each price, not the side, and 1-minute bars approximate them well in one practitioner’s study PlausibleOne trader's study (box below). So delayed or 1-minute data can still serve Module 2’s profiles; it cannot serve this module.

2. Which session template built it? The figure is one session. Across 58 recorded NQ sessions in our lab, the RTH-only and full-session POC landed on the same price in 33; they were 25 points or more apart in 9, 50 or more in 6, and up to 360 points apart (2 Nov 2026) PlausibleNQ. The value-area edges moved by a median of about 19 points. Neither template is “right”: RTH describes the day session most traders act in, the full session includes the overnight inventory (M4.06). Pick one, write it on your level sheet, and never compare a level from one template with a level from the other.

3. How are prints grouped? A “large trade” depends on what counts as one trade. On the same 58 sessions, RTH prints of 20 contracts or more, counted as our recorder stores them, came to about 95 per session. Summing prints that share a timestamp and side (one match event) raised that to about 965 per session; for 50 or more, about 12 became about 137 PlausibleNQ. Same trades, a tenfold difference in “big prints”. Your platform has this setting; know which mode it is in before you read bubbles (M3.11) or the tape (M3.09).

4. Which contract month? A continuous chart switches to the next contract each quarter. In roll week the two months trade a quarter’s carry apart: about 230 points on NQ in September 2026 (M0.08) Definition. A level copied from the old month onto the new one is off by that much, and volume migrates from one contract to the other over several days. Check the month on your chart, your level sheet and your order ticket.

Run the four checks first, settings second, and question the idea last. When a footprint suddenly looks wrong, a stale feed, a changed template, a different grouping mode or a contract switch is a cheaper explanation than a change in the auction, and it is quick to rule out House rule.

The three settings that matter

Everything else can stay at its default. These three should be set from your own instrument and session House rule.

1. Imbalance ratio (M3.03). ATAS X ships at 150%; our desk’s template uses 400%; many courses teach 300%. At 150%, a liquid contract like NQ lights up cell after cell in ordinary two-sided trading. The test is readability: raise the ratio until a session shows a handful of stacks you can actually look at (say 3–8), not every bar. That makes the chart legible. It does not make imbalances predictive: on NQ, single imbalances and stacks showed no edge on their own Lore, so use them only to describe what happened at a level you planned in advance.
2. Delta baseline (M3.04). “Extreme” needs a measured number, and the number changes through the day. On our 58 NQ sessions, the median absolute delta of a 5-minute RTH bar was 173 contracts and the 90th percentile 519. In the first half hour the 90th percentile was 887; between 12:00 and 14:00 ET it was 394 PlausibleNQ. One threshold for the whole day is off by about a factor of two at one end or the other. Measure your own, by time window, and write it down.
3. Large-trade filter (M3.09, M3.11). Fixed tiers such as “20+ contracts is large on NQ” are conventions, not measurements Lore. Use a percentile of grouped trades for your time window, set the grouping mode on purpose (check 3), and review it every quarter, because volatility changes what “large” means.

The meta-rule: a setting is a question about your market, answered by observation. A vendor default is a starting point for the vendor’s average user.

What the data says

Worked example: the trader with 40 imbalances an hour

An illustration, not a real account. A trader’s NQ footprint flags imbalances on almost every 1-minute bar: dozens per hour, all treated as signals, and the account is drifting down. The diagnosis takes two minutes and finds two separate problems.

1. Data. They practise on a replay set to candles. The replay generates the inside of each bar from its open, high, low and close, so the bid and ask columns are invented. Every imbalance on that chart is computed from generated numbers.
2. Setting. The imbalance ratio is the stock 150%, a low bar on a contract as liquid as NQ.

The fix: switch the replay to tick data (M0.07 shows the setting), raise the ratio to 300–400%, and watch a week of sessions. The display drops to a few stacks per session, which a person can read and check against a level written down before the open. The trader did not need a new indicator or a new strategy. They needed real inputs and a readable display, and after the fix they still have no tested edge from imbalances, only an honest picture.

The mirror and the failure

The mirror: filtering too hard. The opposite error hides the market. Set the large-trade filter to 50+ contracts on prints as stored and you see about 12 a session on NQ; group the same prints by match event and the same filter shows about 137. Set the imbalance ratio so high that nothing flags on a trend day and the footprint goes quiet exactly when aggression is heaviest. Calibrate both ends: not every bar, not none.

The failure: tuning the display to your memory. A trader nudges the imbalance ratio until last week’s losing trades no longer show a stack at the entry. The chart now flatters the past and describes nothing. Calibrate to “campaigns are visible, routine trading is not”, set it once, and leave it for a month. And even a well-calibrated display does not turn imbalances into a forecast: in the public study, changing the ratio did not settle the imbalance results (inconclusive), and stacks did not act as support or resistance.

Do it on the platform

Prop-firm note. On this desk you execute in TopstepX and read order flow in ATAS X on your own feed. The two can group and time prints differently, so the firm’s screen is the record for fills and loss limits (MKT2.03, PLAT2.08).

Common mistakes

Common mistake: Optimising settings to make past trades look better. Nudging a threshold until last week’s losers disappear is curve-fitting the display. Set it to describe the market, then leave it.

Common mistake: Treating a calibrated setting as an edge. A ratio that shows four clean stacks a session is a readable chart, not a tested signal. Single imbalances and stacks were tested on NQ and showed no edge on their own.

Common mistake: Copying someone’s level, size tier or delta threshold without their data settings. “POC 30,820” and “POC 30,634.25” were both correct on 1 October 2026; “20 contracts or more” meant about 95 trades a session or about 965 depending on grouping.

Drill: run the four checks on one session

Take one recent NQ or MNQ session on replay or your live chart.

1. Side: confirm the feed records the trade side (M0.07’s five-bar check: ask column minus bid column equals the bar’s delta).
2. Template: build the session’s profile twice, RTH only and full session. Write both POCs and value areas and the distance between them.
3. Grouping: count trades of 20+ contracts in RTH with grouping on and off.
4. Month: write the contract month on your chart, your level sheet and your order ticket. Are they the same?

What we don’t know yet

• Whether ATAS X’s delta equals delta built from CME’s own aggressor flag: not yet measured (queued in our lab).
• Whether the session template or the roll week changes a strategy’s results: inconclusive in the public study, not yet measured by our lab.
• Whether our recorder’s stored prints are single fills, which decides how the grouping counts above should be read: not yet measured.
• Our lab’s own check that 1-minute profiles match tick-built profiles on NQ: not yet measured.

Recap

• Footprints, delta, CVD and the tape need tick data with the side of each trade; profiles can use 1-minute bars (Plausible, one trader’s study).
• Four checks before any theory: trade side, tick vs 1-minute, session template, contract month.
• On our NQ sessions the session template moved the POC by up to 360 points, and print grouping changed the 20+ contract count about tenfold.
• Set the imbalance ratio, delta baseline and large-trade filter from your own percentiles by time of day; they describe the market, they do not forecast it.

Further reading

• Order Flow, Measured: All claims. A public, pre-registered practitioner study on NQ, 2018–2026. Read the measurement rows (ME-01, ME-04, ME-05) and CF-03 on the roll week.
• Lee and Ready (1991), “Inferring Trade Direction from Intraday Data”, Journal of Finance 46(2): why a guessed trade side can differ from a recorded one.
• Our lab: on 58 NQ sessions, the session template moved the POC on 25 days and the print-grouping mode changed the large-trade count about tenfold; neither setting was tested for an edge.

Next: M3.14 asks which tools you can audit at all; this lesson’s checks are the first step of that audit. Earlier links: M3.03 (imbalance ratio), M3.04 (delta baseline), M3.09 and M3.11 (size filters), M0.07 and M0.08 (trade side and roll).

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