Unfinished auctions: three definitions, one cell, and the data underneath

By José Luis Rodríguez Author of ORDER FLOW — Market Mechanics

Read enough about unfinished auctions and you will find the same footprint cell described as finished in one place and unfinished in another. Neither source mentions that the other exists.

The mechanics of the extreme cell are settled; the convention is not. Between the two sit three things that change the cell before anyone reads it: where the data comes from, how many ticks share a row, and which extreme you are looking at. The data source matters most.

What the extreme cell records

A bar reaches its high by going up. Aggressive buyers lift the offer, and the last price that trades is the high. That leaves volume in the ask column of the top cell. The question is what sits in the bid column next to it.

For anyone to sell into a bid at the high, a bid has to exist at the high. That only happens once the offer has moved a tick above it, with passive buyers now resting at the high itself. And nobody can have bought that next tick up, or it would be the high instead. So bid volume at the high means the market sat there long enough to trade on both sides without either side prevailing, and then fell back.

A zero on the bid (0 × N) means the opposite: price turned before the bid ever rose to the high, or nobody hit it when it did. Price found the last aggressive buyer and the rejection was immediate. The low mirrors this, with a zero on the ask marking a closed extreme.

Schematic of the high cell in three different candles. Bid on the left, ask on the right. Values are illustrative.

Three definitions in circulation

The common one. Volume on both sides of the extreme cell is an unfinished auction; a zero on the side that would close the move is a finished one. This is the reading used on this site and in the book, and it is the one most platform documentation describes.

The inverted one. Some guides call the one-sided cell unfinished, reasoning that no opposing aggression ever tested that price. Under that reading the 0 × N cell that most people call closed is the one left pending. The argument is internally consistent, but it runs into the mechanics above: the zero does not mean the price went untested, it means the turn was fast enough that the bid never got there.

The relative one. Some platform documentation does not look for a zero at all. It describes an unfinished auction as an extreme where the buy and sell difference barely changes from the levels below, and draws a different conclusion from it: that price may extend beyond the range instead of coming back.

The three coexist in forums, courses and manuals. Switching platform or source can mean switching definition without noticing.

Tick-direction data cannot show it

A footprint needs to know who crossed the spread on every trade. Real bid/ask data reads it from the quote; the tick rule infers it from price direction, labelling upticks as buys, downticks as sells and repeated prices with whatever came before. The general trade-offs are covered in bid/ask footprints vs tick-direction data. At the extreme cell the error is systematic, as the next paragraph shows.

Within a bar, the high is only reached by going up, so the first trade at the high is always an uptick and gets labelled a buy. Every later trade at that same price inherits the label. For a trade at the high to be labelled a sell, the trade before it would have had to print higher, and nothing inside the bar printed above its high. Under the tick rule, the bid column of the high is zero by construction, and so is the ask column of the low.

Schematic of one sequence of ten NQ trades classified two ways. Highlighted rows are the ones the tick rule labels the wrong way round. Illustrative sequence.

One tape, two opposite cells. Trades seven and eight hit the bid at the high, but they repeat the previous price, so the tick rule counts them as buys. With real data the high is open; reconstructed, it is closed.

To get a sense of scale we ran a simplified order book model (details at the end). With quote data, between 10% and 45% of highs had both columns active, depending on book depth. With the tick rule, under 5% in every configuration tested, and every one of them for the same reason: the first trade of the bar inherited its label from the previous bar.

So on a reconstructed footprint, any unfinished auctions that get flagged are artefacts of where one bar ends and the next begins, and the absence of flags says nothing about whether auctions completed. Before reading extreme cells at all, check which of the two methods the platform uses.

Rows and bars

Ticks per row. Many chart configurations group several ticks into each footprint row for legibility, four ticks being one full point on NQ. The top row then holds the high and the three prices beneath it, where two-sided trade is routine.

Schematic of the same NQ candle at one and at four ticks per row. Aggregated rows sum the levels they contain. Values are illustrative.

At one tick per row both extremes are closed. At four, both look open, because the top row now includes the selling at 21,301.75, 21,301.50 and 21,301.25. In the same model, with four ticks per row, seven to eight highs in ten came out open with either data source. Aggregation erases even the difference between real and inferred data. The extreme cell only means something at one tick per row.

Where the bar ends. A footprint bar closes on time, volume or range, none of which has anything to do with the auction. The high of a one-minute bar halfway up a rally is arbitrary: the next bar opens and keeps climbing, and an open cell there only marks where the clock cut. The reading that carries information is at the extreme of a rotation or of the session, a price the market left and then built structure away from. The book adds conditions of time and distance before treating such an extreme as relevant.

Market Profile describes the same idea at session scale. James Dalton calls the thin tail left by a violently rejected extreme excess, and the flat, tail-less extreme a poor high or poor low. The footprint cell and the profile measure related things with different instruments, and when they disagree neither is right by default. What both try to separate is a move that ends in rejection, where opposing aggression arrives and the bid never settles at the high, from one that ends in exhaustion, where volume fades, delta shrinks without flipping, and the market lingers at the top trading both ways until it falls for lack of buyers.

The threshold question

Assume real data, one tick per row and a well-chosen rotation extreme. One question remains that none of the three definitions settles: how much volume does each side need for the cell to count as open? A 1 × 1 cell has no zeros. Neither does an 8 × 3. Treating them alike seems wrong, and the binary definition does exactly that.

Schematic of the high cells of eight different candles, evaluated against two minimum-per-side thresholds. Values are illustrative.

With the threshold at one contract per side, six of the eight highs are pending. At five, two are, and the 8 × 3 drops out purely because the filter changed. In the model, moving from one contract per side to eight cut the share of open highs from just over one in five to under one in twenty.

The convention on this site and in the book is a cell with no zero on either side. It is a choice, and three questions stay open behind it. Should the threshold be absolute, or relative to the bar's volume, given that a 1 × 1 in a 40-contract bar and an 8 × 3 in a 3,000-contract bar are not the same event? Does an unfinished extreme point to a return, as the common reading holds, or to extension, as the relative one does? And is there a magnet at all? Price revisits recent extremes often, whatever their cells look like. Showing that unfinished ones attract price would need their revisit rate compared against closed extremes at similar distances, on bid/ask data at one tick per row. We are not aware of a published measurement of that kind. The same question hangs over the naked POC, unfinished business at the centre of a session's value rather than at its edge.

Limits and false positives

An open cell describes the past. It says the auction did not finish at that price, without saying whether price will return, when, or what it will do there. What matters is what the flow shows on the revisit.

Trend days. In a session with sustained direction, extremes left open tend to be springboards for continuation. Telling a trend day from a failed breakout while it happens is a separate problem, and the breakout candle alone does not settle it.

Thin books. In the model, book depth moved the share of open extremes more than anything else, with thin queues producing far more of them than deep ones and no extra unfinished business behind the difference. A thinner contract should show the same effect, which is one more reason NQ does not read like ES. That is a hypothesis, not a measurement.

A single contract. Under the binary definition, one one-lot print is enough to turn a closed extreme into an open one. And even a correctly read cell records executions only; the resting orders around the high, and whatever was pulled from it, are outside what a footprint can show.

About the simulation. The figures quoted come from a simplified book with a one-tick spread: aggressive orders of random size consume the opposite queue, the quote moves a tick when a queue empties, and quotes are occasionally cancelled. It was run across 27 combinations of book depth, aggressor persistence and cancellation rate, with bars of 50, 200 and 1,000 trades. It is not NQ data: it compares data sources and settings against each other.