Loading
Two charts, one argument. Everyone publishes a total; a total hides what it rests on. Drawn as bands — read from the contract against reported to us, value moved against claims made — the same figures say something the single line cannot. Counted 10 minutes ago.
On Morpho most of the largest vaults move assets when nobody asked. On Euler, not one does.
A vault that manages money has to move money. An ERC-20 transfer with the vault on one end, in a transaction carrying no deposit and no withdrawal, is the vault acting on its own account — nobody asked it to. That is the same measurement on both venues and it needs no venue-specific configuration, which is the only reason the two are comparable at all.
Euler’s contracts agree with the reading. Every event they emit was named from a verified ABI — VaultStatus, Transfer, InterestAccrued, Repay, Deposit, Withdraw, Borrow, Approval, ConvertFees — and every one is a user acting or interest accruing. There is no operator event on the contract to find.
So this is not a fault. An Euler vault is a lending market: depositors supply, borrowers borrow, and the rate is arithmetic rather than a decision. A Morpho vault has a curator who allocates between markets, and that allocation is what shows up here. The point is that two things sold in the same sentence — “a vault that earns yield” — are doing very different amounts of deciding, and until now nothing said which was which.
Both figures are over the largest vaults we can read on each venue, not all of them: 24 of Morpho’s 155 and 15 of Euler’s 208. Borrows, repayments and liquidations are counted as users, without which every borrower on a lending market reads as the manager acting.
Where a network’s figure is an average across everyone building on it, a venue’s is a decision somebody made. The column on the right says whether that venue’s vaults have been watched acting at all — a separate reading from who owns them, and the one that decides whether the bar can ever grow.
Every published figure for this market is a count — how many agent vaults exist, how many agents registered, how many launched this month. A count treats a vault holding four billion dollars and a vault holding four dollars as one each, and that is the arithmetic a venue is rewarded for.
We hold 213 vaults here and no capital reading for any of them, so there is nothing to divide. That is a gap in what we read, not a network with nothing in it.
Three different things wear this label and they are kept apart above. The dark band is read from what the vault’s own transactions did on chain. The middle band is a venue listing its own product as agent-operated, taken at face value and marked as such. The pale band is a vault whose NAME contains an autonomy word — “ai” is a substring of ordinary English, and that band exists to be visible and small rather than to be believed.
Each vault counts in one band only, its strongest evidence. A vault can carry several, and summing the methods would count its capital once per method — bands that overlap cannot be stacked, and a chart whose segments exceed their own total is worse than no chart.
Absent entirely: Drift (213 vaults) — we hold the vaults and have read no capital for any of them. That is a gap in our reading, not a network holding nothing, and a zero there would have been the opposite claim.
Every capital chart in this market is one line going up, and it never says where a dollar came from — whether somebody read the contract, or a venue reported its own book, or an aggregator was asked and its answer inherited along with its staleness. Same total, three strengths of evidence.
This is a cohort, not the market. Only the 162 vaults with a reading in every one of the last 6 months are in it, so the level is below the site’s headline total and the shape is real. The first version of this chart summed every vault observed each month and rose from $330M to $5.19B — almost all of which was our own coverage improving, not capital arriving. A capital chart whose slope is the reader getting better is the most misleading thing this site could publish.
Every vault in this cohort is the venue's own book. Not one is read from its contract, at any window from three months to twelve — so there is no contract-against-reported split to show here yet, and no switch offering one. An option that filters to an empty chart reads as a broken page rather than as a fact about coverage.
The 392 vaults we do read from their contracts average 5.5 months of share-price history and the deepest has 9, none of it continuous. That is the gap: the capital we can verify is the capital we have least history for. The split returns when the backfill has given those vaults a continuous series, which is work rather than a label change.
Bands are attributed by how each vault is read today, carried backwards: source was not recorded per observation until this week.
The site’s founding question drawn as a shape rather than as today’s percentage. Every other figure here describes a moment; this one says whether the moment is a trend.
Same cohort as above and for the same reason: summing every vault observed each month would draw our coverage improving and call it agents taking over. Both bands are equally well measured here, so neither is a claim against the other — the question is about the world, not about our confidence in it.
A registration says an agent exists. A settled job says one was paid. Both are real events on real chains and only the second is evidence of an economy — so adding them together is not a summary, it is the question being averaged away.
Both charts are built from records this site read itself, and both are drawn from the same component so the two cannot drift apart in how they treat evidence. What each figure rests on is in Methodology.