Five ways a large position gets built, and what each leaves on the tape

Large positions are not placed the way small ones are. Somebody moving size into a market with a thin book has to decide how much price impact to accept, how visible to be, and over how long — and those decisions leave different traces.

Those traces are worth learning to recognise. They are also worth learning the limits of, which is the part that usually gets left out: every shape below is consistent with at least two completely different intentions, and the tape does not disambiguate them. This article covers both halves.

The counts below come from our own store, all measured on 17 August 2026, and one filter applies throughout: trades of $1,000 or more and up to $10 million, with an identified market. The ceiling is there because a thin tail of recorded amounts above it is not credible, and leaving them in would move every total we quote. That leaves more than 1.1 million trades, from more than 94,000 addresses across more than 44,900 markets, captured between 2 February and 17 August 2026. Smaller trades are not collected, so every figure here counts size rather than everything — which is itself part of the argument.

1. Slow accumulation

What it looks like. The same address appearing repeatedly in one market over hours or days, with consistent directional bias, each individual trade small relative to the total. Twenty modest orders rather than one large one.

That last shape is the rare one. Grouping the buy side of the store by address, market and outcome — YES or NO, not buy or sell — gives more than 276,000 builds; about 208,000 of them — 75.25% — are a single trade, and about 4,100, or 1.49%, run to twenty slices or more.

Why anyone does it. Splitting an order is the standard answer to price impact. Each slice consumes a little of the resting book, which refills between slices; the total fill is far better than crossing the whole size at once would have been. It is the same reasoning behind scheduled execution in any market, applied by hand.

What it does not tell you. Whether the position is finished. Accumulation is only identifiable in retrospect, because the shape of "someone building patiently" and the shape of "someone who bought three times and stopped" are the same shape until one of them continues. It also does not distinguish a directional build from a market maker accumulating inventory it fully intends to lay off.

2. The single large print

What it looks like. One trade that moves the price on its own, usually accompanied by a burst of volume in a short window.

These are rare and they are large relative to what is resting. In the store, roughly 1,900 trades are $100,000 or more — 0.167% of the total — and about forty are a million or more. The largest single print under our $10 million ceiling is $5,372,553.40, bought at 0.999 at 01:49 UTC on 17 April 2026. For scale, the median order-book snapshot we hold carried about $44,500 of recorded ask-side depth when we measured it, at 10:39 UTC on 17 August 2026, and more than half of the snapshots we hold carry less than $100,000 — more than 11,400 snapshots, running from 16:28 UTC on 15 August to 10:39 UTC on 17 August 2026. An order of that size is not a fill; it is an event.

Why anyone does it. Because speed was worth more than price. If the information is about to be public, the impact cost is cheaper than the risk of being late — a straightforward trade-off, and the visible impact is the price of taking it.

What it does not tell you. Almost anything about conviction. A single large print is equally consistent with someone unwinding in a hurry, with a hedge being put on against an exposure held elsewhere, with an inventory adjustment, or with an error. It is the loudest event on the tape and one of the least legible.

It is also the one most likely to have a bad fill waiting behind it for anyone following: the print consumed the book, and everyone reacting to it arrives into what is left. That mechanism, and six others, are in seven ways copying a wallet goes wrong.

3. The scale-in against the move

What it looks like. Repeated buying as the price falls away from the first entry — average price improving, position growing, direction unchanged.

It is the minority case. Of more than 36,900 builds of three slices or more, 27.01% ended at a price below their first slice; about 23,000 ended higher and about 4,000 ended flat. Most repeated buying chases the move rather than fading it, which is worth knowing before you read a scale-in as the default explanation for a wallet that keeps appearing.

Why anyone does it. If your probability estimate has not changed and the price has moved against you, the contract is now better value than it was when you first bought it. Adding is the arithmetically consistent response, and on a bounded payoff the improvement is calculable rather than hopeful.

What it does not tell you. Whether the estimate is any good. This shape is identical to a trader refusing to update, and the fact that a position is bounded below at zero is exactly what makes averaging down survivable right up until it is not. On the crypto side of our coverage the same behaviour with leverage attached is what builds a liquidation cluster — the mechanism is in liquidations and cascades.

4. The quiet exit

What it looks like. Very little. Sales distributed over time, frequently posted as resting liquidity rather than crossed, sized to be absorbed by the flow already present.

The store shows the gap plainly. Of more than 255,000 address-and-market pairs with at least one buy of $1,000 or more, only 8.47% ever show a sell of that size. Across the whole store, buys outnumber sells by nearly three to one — more than 838,000 against more than 296,000. Some of that is the $1,000 threshold cutting exits into pieces too small to record, and some of it is that leaving a binary market is often done by buying the other side rather than selling this one. Both causes point the same way: the exit is the half you do not see.

Why anyone does it. The same impact logic as accumulation, in reverse, plus a second motive: an exit that is recognised as an exit is expensive, because it invites others to leave first.

What it does not tell you. That it is happening. This is the important asymmetry of the whole article: builds are legible and unwinds frequently are not. Anyone reading flow will systematically see more entries than exits, which biases every impression they form about what a tracked address currently holds. Assume you are late on exits and you will be right more often than not.

5. The hedge that looks like a reversal

What it looks like. An address that has been buying one side starts buying the other, in the same market or in a correlated one.

Common enough to matter: more than 21,000 of those same address-and-market pairs — 8.35% — bought both YES and NO in the same market.

Why anyone does it. Because the exposure is being reduced rather than the view being changed — or because the other side has become cheap enough that owning both is the trade, which on a bounded payoff is a real position rather than a contradiction.

What it does not tell you. Which of those it is. And on a binary contract this shape is harder to read than anywhere else, because buying NO and selling YES are the same transaction: what appears as a new position on one book is a reduction on the other. What a prediction-market order book does that a crypto book does not covers that mirroring in detail, and it is the single most common source of misread flow on these venues.

What the tape is actually good for

Read the list again and notice the structure: every pattern has a plausible mechanical explanation and at least one alternative explanation that produces the same observation. That is not a defect in the observation. It is what flow data is.

So the useful posture is narrow:

• —Patterns tell you about execution, not intent. They describe how someone chose to trade against the liquidity available. That is a real, measurable thing, and it is a different question from why.
• —A pattern is a prompt to read the market, not a position. The thing worth doing after noticing an accumulation is reading the resolution rule and forming your own probability.
• —Never infer conviction from size. The population of large addresses is drawn from a population that, in aggregate, loses. Of the 2,448 addresses in our operator ranking — each with at least twenty closed trades, recomputed 17 August 2026 — 1,585, or 64.75%, are down on realised PnL, with a median of −$225.77. Bloomberg reported on 28 April 2026 that most prediction-market traders were losing money: most prediction market traders are losing money while bots rack up gains.
• —Never infer a track record from a pattern. Whether an address is worth reading at all is a separate measurement with its own window, denominator and filters — see what "smart money" means, and what it does not, which goes through what happened when we audited the obvious version of that ranking on our own store.

The counts above are written as orders of magnitude for a reason, and the two that are not carry the minute they were read. Re-run the same queries an hour later and the totals will have moved, because the store keeps ingesting; the shares — three quarters of builds being a single trade, 8.47% of positions ever showing an exit — are the stable part, and they are the part worth carrying around.

We publish an operator's activity and the depth of their history because reconstructing it yourself is a genuine data problem, not because the reconstruction settles anything. It narrows where to look. The judgement stays yours, and the only feedback loop that will ever tell you whether your reading of these shapes is any good is your own trade log, compared against the prices you paid.

Related reading: what funding rates say about positioning, for how the same question — where is exposure concentrated, and what happens when it unwinds — is answered on the crypto side, where the positioning is directly measurable instead of inferred.