Where the leverage is

How many of these traders we can name

Every address we hold — declared by an operator or recorded in an ERC-8004 registry — asked against every venue that can answer an address join. The reads are complete; what they find is almost nothing.

Venue	Addresses we hold	Found trading	Why that is the number
GMX	980	0	Every open position on both chains was read and joined by address. A GMX position is held by the trader’s own key, so a join works here — it simply finds nothing.
Ostium	943	0	Ostium trades from the trader’s own Arbitrum address rather than through a proxy, so an address join works here where it cannot on Polymarket. A zero means no address we hold trades on Ostium.
Polymarket	113	3	Polymarket trades through PROXY wallets created by the exchange, not the key an operator declared or a registry recorded. An agent can be active here and invisible to an address join. Zero means no declared address is a Polymarket proxy — never that no agent trades here.
dYdX	—	0	dYdX accounts are Cosmos addresses. No declaration, registry or rail we read carries one, so no dYdX account is attributable to an agent. The reader exists and is called with an empty list.

The positions standing on the least of their own money

Collateral plus unrealised profit, over size. Not a liquidation distance — the real threshold is a per-market maintenance factor this does not read, so a thin buffer means a position with little of its own money left, not one about to be closed. Positions above $50.0k only.

Three costs, kept apart, because they answer different questions: a fee is charged for trading, funding is paid to whoever is on the other side, and borrowing is rent on the position. A venue quoting only the first is quoting a third of what it costs to be there.

Of 22,354 orders that set a limit price, 99.9% filled inside it — which is not news, because the venue enforces that. The informative figure is the room they had: 99 bps of margin against their own limit at the median. 10,186 more set no limit at all and are counted apart rather than scored as perfect.

The other leverage: what agents owe

6,469 attributable addresses asked directly, across Aave V3 and Morpho Blue. 100 positions in total, posting $994.5k and borrowing $40.2k against it.

Per protocol: Aave V3 92 positions (read from the chain); Morpho Blue 8 positions (the protocol’s own indexer).

Agents post collateral and do not borrow against it. That is the finding, and it is the opposite of what the leverage on the venues above would lead a reader to expect. It is also narrow on purpose: Aave V3 and Morpho Blue are read and Compound, Euler, Spark, Fluid are not, so a zero means no position on the protocols we ask rather than no debt anywhere. Every chain in scope answered.

Two leverage figures, and why this page publishes ours

GMX’s indexer carries a leverage field of its own. Read at the scale that makes it comparable, it lands within a quarter of our figure on 488 of 915 positions and outside it on the rest. The median ratio between the two is 1.03, with a tenth of positions below 0.61 and a tenth above 2.00.

A median near one with a spread that wide is not a scale error — a wrong scale would be wrong on every position by the same factor. It is a different definition, and the likeliest one is that the venue divides by equity where this divides by collateral: a position deep in profit or loss is exactly where those two part company. Ours is published because its units are established from the venue’s own price convention; theirs is described here because a reader comparing this page to GMX’s own interface deserves to know they differ.

One vocabulary for five venues that share none. Every square says which of four things a blank would otherwise hide — and only two of the four are ours to close. 85% of the fields that can be filled are filled, 19 of 22 fields are readable on at least one venue, and 3 are readable on none.

Field	GMX	dYdX	Ostium	Avantis	Polymarket
Account	●	×	●	×	●
Venue	●	●	●	●	●