"A full set of outcome tokens is always worth $1, so assembling one for less is free money." This is the most obvious arbitrage on the venue and the one I get asked about most. I censused the whole lane: 694 participants, zero live executable inventory — and 39% of the lane loses money.
Every mutually exclusive outcome has its own token, and the rule is simply the winner pays $1 and everything else pays $0. So regardless of the result, a full set is worth exactly $1.
All true — the mechanism is real. The question is not the mechanism, it is how much is left.
| Question | Measured |
|---|---|
| Participants in the lane | 694 |
| Live executable inventory | zero |
| Share of the lane losing money | 39% |
| Share that is genuinely riskless | only 22% |
"Zero inventory" is roughly what you would expect. But 39% of participants losing money does not fit the phrase "riskless arbitrage" — nothing riskless should produce losers.
The fourth row explains it: of the activity in this lane that gets called arbitrage, only 22% is genuinely riskless. The rest carries exposure of some kind — and the people running it may not realise.
You see the outcomes summing to 0.96, four cents sitting there. It usually fails one of these:
| Gate | Detail |
|---|---|
| ① Quotes are not executable size | those four cents typically live in a band with a few dozen shares, or that cannot be filled at all |
| ② Every leg pays a fee | N outcomes means N taker fees, not one |
| ③ Unpaired is not arbitrage | a set needs N fills; one missing leg and the set does not exist |
Gate three is the lethal one: a missing leg is not "slightly less profit", it is a naked directional position — and you entered this trade specifically to avoid direction. That is almost certainly where the 39% comes from.
On a multi-outcome board the near-zero-probability bands usually have no quotes at all. And you must buy them to complete a set.
There is a compounding constraint: the matching layer's minimum of 5 shares applies to every leg. More outcomes means a larger minimum viable trade — so cold bands are either unbuyable or force you to oversize. See minimum 5 shares.
Multi-outcome boards offer conversion: "NO on one outcome" and "YES on all the others" are logically identical and can be swapped.
The usual hope is that conversion routes around the taker fee. It does not. Leg fees are paid when you build the position; conversion is a chain action that only changes when cash is realised.
It has value — better capital turnover — but that is time value, not a fee discount. Modelling "conversion saves fees" gives a systematically optimistic result.
And in practice: many toolchains have no conversion entry point at all. I measured this — the SDK in use offered split, merge and redeem only. Confirm you can perform the action before you go looking for the opportunity.
So the correct default for this class is "assume it is already taken". If you can still see it, that is usually because it cannot be executed, not because nobody noticed.
The practical value of that judgement: it saves you from building infrastructure for weeks before discovering there is nothing there. Counting what remains takes a day.
The mechanism is real and the inventory is empty. More striking: 39% of the people in this "riskless arbitrage" lane are losing money, because only 22% of it is genuinely riskless — the rest are carrying naked-leg risk without knowing it.