Market Maker vs Wash Trade: Where Chain Data Stops
Two-sided trading by a small set of related wallets fits a market maker and fits an operator producing turnover equally well. This page separates the features that genuinely lean one way from the features that only look decisive, sets out the cases where the record cannot choose, and shows how this desk words a conclusion when the evidence stops short of one.
- Question
- Can on-chain data tell market making apart from deliberately produced turnover?
- Evidence used
- Swap and liquidity events, pool depth over time, inventory paths, buy and sell value balance, fee spend and activity envelopes.
- What it cannot show
- Intent. The record has no motive field, and both activities are legal, ordinary and openly transacted.
- Falsified by
- A case where depth rises, inventory drifts and adverse selection is absorbed, yet the operator states the flow was produced to order.
- Confidence
- Moderate. Several features lean reliably, but a competent overlap of the two behaviours remains undecidable from the record alone.
A market maker and an operator producing turnover can leave nearly the same trail. Both trade both sides, both trade often, and both work from a small set of funded addresses. The separating fact is intent, and intent is never written to the chain. The honest output in this zone is a leaning with a stated confidence, not a verdict.
Why both readings fit the same trail
Market making is the business of standing ready to buy and to sell the same asset, earning the spread while carrying inventory risk. Producing volume is the business of generating turnover so that a pair reads as more active. Reduced to a list of swaps, both collapse into one object: repeated two-sided trading by a handful of related accounts.
The overlap is not an edge case, it is the normal state of the data. A transaction record carries accounts, amounts, program invocations, slots and fees. It carries no motive field. Anything said about why a wallet traded is inference layered on the record, and it has to be presented that way.
The comparison is worth making at all because turnover generation is an openly sold service rather than a hidden practice. Teams buy it the way they buy a listing or a design retainer, and vendors publish what the category does, so a professional Solana volume bot is a documented product rather than a secret. Treating the category as ordinary is what lets an analyst read it calmly.
What a market maker actually does on chain
On an order book venue the mechanics are visible as intent. A maker posts a bid and an offer around a mid price, cancels and replaces both as the reference moves, and trades only when somebody crosses the spread. The chain records the quote updates as well as the fills, so passive presence is itself an artefact that can be counted and timed.
On an automated market maker there is no resting quote to observe. Liquidity sits in a pool, the curve prices it, and a participant expresses a view by adding or removing pool positions and by swapping to rebalance inventory. What the record shows is a series of swaps and liquidity events. The passive half of the job leaves a far fainter trace, and that faintness is the source of most ambiguity on Solana venues documented by Raydium and routed through aggregators such as Jupiter.
| Venue type | What the record shows | What stays invisible |
|---|---|---|
| Order book | Quote placement, cancellation, replacement rate, resting size at each level, fills | Why a quote was pulled, and whether a fill was wanted |
| Pool based | Swaps, liquidity added and removed, range choices, pool balances after each event | All passive quoting, since the curve quotes on the provider's behalf |
| Aggregated route | The final settled legs across whichever pools were touched | The intent behind splitting, and which leg the trader cared about |
Pool venues therefore delete the most informative artefact a maker produces, the standing willingness to trade at a price nobody has taken yet. On a book that willingness is measurable directly. In a pool it can only be inferred from depth, and depth belongs to everyone who supplied it.
Features that lean towards market making
Five features carry real weight. Two-sided presence around a mid price that is itself moving, rather than around a fixed number. Inventory that drifts away from flat and is then rebalanced deliberately. Willingness to be adversely selected, meaning fills that lose money when the price runs. Depth that increases while the participant is active. Quotes or swaps that respond to price moves occurring somewhere else first.
Adverse selection is the strongest of the five, because it is the one that costs money to fake. A quoting participant collects small spreads most of the time and loses on the fills that land just before a sharp move. That loss pattern is the fingerprint of real risk, and its absence is informative without proving anything.
Observation: constant two-sided activity
Read as market making, constant presence on both sides is simply the job. A maker cannot choose when the other side arrives, so it must quote through quiet hours and busy ones alike, and its trade count stays high regardless of what the price is doing.
The supporting detail is the mid price. If the two-sided activity recentres itself as the reference price moves, the participant is tracking a market rather than filling a schedule.
Observation read the other way
Read as produced flow, constant two-sided activity is what a session looks like when the objective is a turnover figure. Buying and selling in alternation is the cheapest way to accumulate volume without taking a directional position that has to be funded.
The supporting detail is again the mid price, but inverted. Activity that continues at the same cadence while the reference price moves sharply suggests a schedule that is not reading the market at all.
What would falsify this
If a participant showed all five market making features and the operating team then described the activity as a purchased turnover engagement, the feature set would be shown to be reproducible rather than diagnostic. That outcome is possible. A sufficiently well designed programme could add depth, hold inventory and accept losing fills, and the reading here would have to be downgraded across every case that relies on those features.
Features that lean towards produced flow
Four features lean the other way. Turnover that rises with no depth added anywhere. Buy value and sell value that are near symmetric over short windows rather than long ones. An activity envelope that starts, runs and stops as though a budget were being spent. And an absence of adverse-selection cost, meaning the participant somehow never holds the losing side of a fast move.
Symmetry deserves care. Market makers are inventory neutral by design over long horizons, so symmetric totals are expected in both readings. The distinguishing element is the path between the endpoints, not the endpoints themselves. A maker accumulates, carries and unwinds. Flow designed to produce turnover tends to return to flat far more often and far sooner than risk taking would require, a pattern examined in detail under round trips and self-trading.
Observation: buy value equals sell value
Read as market making, a balanced book is the desired end state. The maker started flat, took inventory in whichever direction the flow pushed it, hedged or unwound, and finished flat again. The balance is the result of a risk process, not its purpose.
What supports this reading is a visible middle. Inventory should be measurably away from zero for meaningful stretches, and the size of that excursion should correlate with how one-sided the market was.
Observation read the other way
Read as produced flow, a balanced book is the design constraint rather than the outcome. Holding inventory costs capital and creates price exposure that nobody buying turnover wants, so the cheapest configuration returns to flat almost immediately after each excursion.
What supports this reading is a missing middle. Inventory that never departs far from zero, and departs for seconds rather than hours, describes a process built to avoid risk rather than one built to price it.
Illustrative arithmetic, invented round numbers describing no real pair
Take a hypothetical pair called PAIR-X over one day. Suppose 400 swaps settle, averaging 5 SOL each, for 2,000 SOL of turnover, with buy value at 1,010 SOL and sell value at 990 SOL, and pool depth reading 300 SOL at the start of the day and 300 SOL at the end.
The fee side is a protocol fact: at a base fee of 5,000 lamports per signature and one signature per transaction, 400 transactions cost 2,000,000 lamports, or 0.002 SOL. Turnover of 2,000 SOL was produced against a trivial fee outlay, with no change in the depth available to anybody else. That combination leans towards produced flow. It does not prove it, because a maker rebalancing an existing position would spend the same fees and add no depth either.
The leaning table
The table below is the working form of this article. Each row takes one observation, states the case for both readings, and names the fact that would settle it. Nothing in the final column is always obtainable, which is why several rows stay leanings.
| Observation | Leans market making because | Leans produced flow because | What would decide it |
|---|---|---|---|
| High trade count, both directions | Continuous quoting produces exactly this shape | Alternating sides is the cheapest way to add turnover | Whether the recentring follows an external mid price |
| Depth unchanged while turnover triples | A maker can rebalance without adding new liquidity | Turnover was the objective, availability was not | Liquidity events by the same cluster in the same window |
| Inventory returns to flat within seconds | Tight risk limits can force very fast unwinds | Carrying risk is a cost the buyer of flow does not want | The distribution of holding times, not the mean |
| No losing fills around sharp moves | Good latency and a fast cancel policy can achieve this | Trading against yourself never loses to informed flow | Counterparty identity across the whole cluster |
| Activity starts and stops on a clean boundary | Mandates begin and end, and capital is reallocated | Budgets are consumed and sessions terminate | Whether the boundary coincides with a reward window |
| Trade sizes drawn from a narrow set | Risk limits quantise order size in most maker systems | Configuration fields quantise size the same way | Whether size varies with observed book pressure |
Where the reading stays undecidable
Some cases are not hard, they are undecidable from the record. A participant that adds depth, holds inventory for minutes, occasionally loses to informed flow and also happens to be paid per unit of volume is doing both jobs at once. There is no partition of the swap list that assigns each trade to a motive, because the motives are not disjoint in the first place.
The second undecidable class is the well capitalised operator that deliberately absorbs some adverse selection. This desk does not publish guidance on making produced activity harder to detect, so the point stops at acknowledgement: the possibility exists, it degrades the strongest feature in the list, and any reading resting on that feature alone should say so.
Moderate rather than High because the decisive feature, adverse-selection cost, can be reproduced by a participant willing to spend on it, and because pool venues hide passive quoting entirely. Moderate rather than Low because depth, holding-time distribution and envelope shape agree with each other far more often than chance would allow, and three agreeing weak signals are worth more than one strong assertion.
What would falsify this
If holding-time distributions, depth changes and envelope boundaries were shown to disagree with each other as often as they agree across a large sample of pairs, the combined reading would have no more value than a coin toss and the whole framework would need rebuilding from a different feature set. Anyone testing that should sample by liquidity band rather than by suspicion.
Volume-linked incentive programmes
When a venue, a chain or a listing venue keys rewards to traded volume, it changes the economics of both activities simultaneously. A genuine maker can accept a wider loss on the spread because rewards subsidise the position. An operator producing turnover gains a direct, calculable payoff. The two behaviours converge under incentives, and the correct response is to widen the confidence band during those windows.
Incentive windows also create a timing artefact worth logging. Activity that begins on the first slot of a qualifying period and ceases within minutes of its end is responding to the programme, and both a maker chasing rewards and a produced-flow session chasing the same rewards will show that boundary. It rules out coincidence, not motive.
None of this speaks to the operator's own exposure, which is a separate question with a separate answer. A team that hands funding keys, wallet control or signing authority to an external service is accepting counterparty risk that has nothing to do with market structure, and the practical questions around volume bot safety concern custody, key scope and withdrawal control rather than how the resulting trades read on chain. An analyst should keep the two apart.
Wash trading is a term with a legal home
Wash trading has a specific meaning in regulated markets. It describes trades arranged so that beneficial ownership does not really change and no market risk is transferred, and it is treated as a form of deception because the resulting print misleads other participants. Reference definitions are maintained by bodies such as the CFTC and summarised in general references on wash trading.
This desk does not apply that term to any identifiable party. It has legal weight, it implies a finding about intent, and no volume of chain data supports such a finding on its own. Where a pattern matches the shape, the desk writes that it is consistent with produced flow and states the confidence. Naming a token, team or launchpad as a wash trader is an accusation the method cannot carry.
The vocabulary matters for a second reason. Produced volume is a normal, openly transacted category. Collapsing it into a regulatory term borrowed from another jurisdiction destroys the distinctions an analyst is trying to preserve.
Questions to ask before any conclusion
Before writing a sentence in this zone, work through the list below. If any answer is unknown, the conclusion has to name it as unknown rather than assume it away. The banding rules for the resulting statement are set out under stating confidence honestly, and the shared terms are defined in the glossary.
- Did depth available to other traders change while this participant was active?
- What does the distribution of inventory holding times look like, not just its average?
- Are there fills that lost money immediately after execution, and how many?
- Does the activity recentre on a moving reference price, or repeat around a fixed one?
- Does the start or the end of the envelope coincide with a reward or listing window?
- Would a competent market maker with a small mandate produce this same record?
- Which single observation, if reversed, would change the conclusion?
- Is any part of the statement an accusation the record cannot support?
The output that survives this list usually reads narrower than the analyst expected, and that is the point. A conclusion in the grey zone names the observation, gives both readings, says which way the evidence leans and by how much, and ends with the fact that would overturn it.
Questions this desk is asked
Can on-chain data prove that a wallet is wash trading?
No. The record shows accounts, amounts, program calls, slots and fees. It does not show why anyone traded. Wash trading is defined by intent, specifically trading with no change in beneficial ownership or market risk, and intent is not a field you can query. The strongest honest output is a leaning with a confidence band attached, plus the observation that would overturn it.
What single feature separates the two most reliably?
Depth. A market maker exists to make a market deeper or tighter, so its presence should show up as liquidity that is available to others, not only as completed swaps. Turnover that rises while pool depth and resting size stay flat leans towards produced flow. It is a lean rather than a proof, because a maker can rebalance actively without adding a single unit of new depth.
Does symmetric buy and sell value mean the flow is produced?
Not on its own. Near symmetric value over a window is what inventory-neutral trading looks like, and market makers are inventory neutral by design over long horizons. What matters is the path between the endpoints. A maker accumulates a position, carries it at risk, and unwinds it later. Produced flow tends to return to flat far more often and far more quickly than risk-taking would require.
What is adverse selection and why does it matter here?
Adverse selection is the cost a quoting participant pays when better informed traders trade against a stale price. A genuine maker absorbs that cost as the price of collecting the spread, so its record contains losing fills around sharp price moves. A pattern that never loses to informed flow is either avoiding risk deliberately or trading with itself, and both readings deserve to be stated.
Do incentive programmes change how this should be read?
Yes. When rewards, points or listings key off traded volume, both activities gain a reason to trade more, and the economics of each change at once. A maker widens its acceptable loss because rewards subsidise it. An operator producing turnover has a direct payoff to target. Incentive periods therefore compress the gap between the two footprints and should widen, not narrow, the confidence band.
Why does this desk avoid naming parties?
Because the term wash trading carries a specific meaning in regulated markets, and applying it to an identifiable team or token would be an accusation the record cannot support. The desk describes patterns, states what each pattern leans towards, and stops there. Anonymised placeholders and hypothetical pairs let the method be examined and reproduced without attaching a claim to anyone in particular.
What does a fair conclusion in this zone sound like?
It names the observation, gives both readings, says which one the evidence leans towards and by how much, and states the fact that would flip it. A usable form is: turnover in this window is consistent with produced flow at moderate confidence, because depth did not change while volume tripled, and a record of quotes absorbing informed flow would overturn that reading.
Filed in Grey zone by The Volume Forensics Desk. Patterns described here come from protocol design and public transaction data; every figure in an example is invented, labelled and describes no real pair. The evidence standard the desk works to is set out in stating confidence honestly, and the terms used are defined in the glossary.