Loading
Evidence
What every number here rests on, in the order a claim travels.Note 22
Capital you can check without trusting us
read 2026-10-10$10.53B across 733 vaults, where the quantity the contract answers with, multiplied by the price we publish, lands within two per cent of the figure we publish. Both halves are checkable: the quantity by anybody with an endpoint, the price beside it. That is 95% of the capital in vaults that hold a contract we can ask.
17 do not agree and are listed rather than quietly excluded. A gap here has four possible owners, our price, our reading, the vault not being ERC-4626, or the venue, and this check corrects none of them. Writing a corrected figure from it would replace one unverified number with another and call it proof. 1,772 could not be asked, holding $493.93M between them ,$294.59M because the venue is an order book, so no contract holds the capital to be asked; $73.43M because a share token whose accountant last set its rate 26 hours ago; $27.18M because a share token whose vault names no accountant contract to ask for a rate; $25.43M because the vault answers neither totalAssets() nor totalSupply(); $25.39M because we know which token it is and hold no price for that token, and its ticker belongs to more than one asset; $21.96M because a vault whose own accounting snapshot is 72 hours old, which is not a figure for now; $8.79M because a vault whose own accounting snapshot is 39 hours old, which is not a figure for now; $7.07M because a vault whose own accounting snapshot is 269 hours old, which is not a figure for now; $4.33M because no price for the denomination, and no address to ask about instead; $2.90M because a share token whose accountant last set its rate 3699 hours ago; $2.59M because totalAssets() answered with nothing usable; $166.7k because a share token whose accountant last set its rate 338 hours ago; $116.5k because the job ran out of budget before reaching them.
The same claim, at a block you can name
read 2026-10-09The figures above were asked at the head of each chain and the block was not recorded, so nobody can repeat them. These can be repeated. Each row is one call, at one block, and the answer is stored exactly as the contract gave it. 38 of 60 land within one per cent of the figure we publish elsewhere on this site.
4 of those land on our figure to the last digit, which is the weaker result rather than the better one. It means our stored number came from the same quantity and the same price, so there was one reading and not two, and the row proves the chain was read correctly without proving our figure is current. The rows that agree to within a fraction of a per cent are two readings taken at different moments landing in the same place.
The vaults we currently disagree with are listed first, inside the rows that are free, so the sample cannot be one chosen by its own result. A reader can re-run a row where our figure and the venue's part ways, at the block named beside it, and reach their own verdict.
Each row is one eth_call. POST {"jsonrpc":"2.0","id":1,"method":"eth_call","params":[{"to":CONTRACT,"data":CALL},BLOCK]} to any RPC endpoint for that chain, with BLOCK as the hex of the block field. The result is the raw field here, verbatim. Divide it by ten to the decimals to get the quantity, multiply by the price to get the dollars. The quantity is the chain’s and you can check it; the price is ours and is named beside it. The endpoint has to serve state at a past block, since the block named here is older than the moment you read this: ethereum-rpc.publicnode.com and eth.drpc.org both do, free, and neither is ours.
Re-read through endpoints that are not ours. Not every free node serves state at a past block, and the best-known one refuses. These answered with the same figure we publish: avalanche (avalanche-c-chain-rpc.publicnode.com, api.avax.network/ext/bc/C/rpc); base (base.drpc.org, base-rpc.publicnode.com, mainnet.base.org); berachain (berachain.drpc.org, rpc.berachain.com); bnb (bsc-mainnet.public.blastapi.io, 56.rpc.thirdweb.com); ethereum (eth.drpc.org, eth-pokt.nodies.app, rpc.mevblocker.io); monad (rpc.monad.xyz); sonic (sonic.drpc.org, rpc.soniclabs.com); world (worldchain-mainnet.g.alchemy.com/public, 480.rpc.thirdweb.com).
On some chains no free endpoint would serve the block, so checking those rows needs your own archive access. Every candidate we asked refused, and the reasons were pruned state, a token requirement, or no support for the call at all: hyperevm. The rows are no less true; what is missing is a free way for somebody else to confirm them.
An outside endpoint disagrees with our reading. hyperevm: https://rpc.hyperliquid.xyz/evm answered 0x00000000000000000000000000000000000000000000126fb94dd902d7c2c14e where we read 0x00000000000000000000000000000000000000000000126fb94b66c9b2832fc3, reproducibly: asked twice for the same block it gave the same answer both times.; hyperevm: https://hyperliquid.drpc.org answered 0x00000000000000000000000000000000000000000000126fb934d241a24158c3 where we read 0x00000000000000000000000000000000000000000000126fb94b66c9b2832fc3, reproducibly: asked twice for the same block it gave the same answer both times.. Two nodes reading one contract at one block cannot both be right, so this is either their node or ours, and it is published before it is explained.
| Vault | Chain | Block | What the contract answered | Quantity | Price | On chain | Ours |
|---|---|---|---|---|---|---|---|
| EVK Vault exUSD-2 0xf1ba8c5c…acb9eb · 0x01e1d114 | ethereum | 26,156,800 | 0x00000000…9158d76f3e | 12,718,888.6772 XUSD | $1.265697363 | $16.10M | $16.10M agrees |
| Concrete · CTDELTAWEETH 0xb9dc54c8…ee8f20 · 0x01e1d114 | ethereum | 26,156,800 | 0x00000000…460a5145ac | 278,170.834213 WEETH | $2,840.145573 | $790.05M | $790.05M agrees |
| Gauntlet USDC Prime 0xee8f4ec5…e44b61 · 0x01e1d114 | base | 52,391,503 | 0x00000000…ece8c4e523 |
Analyst tier
20 of 60 shown in full. 40 more are listed with the vault and contract withheld. A plan opens every vault in the proof sample with its contract, the block it was read at, and the integer the contract answered. The count is free because the count is the finding; the detail is what Analyst carries.
Where our withdrawable figure and the contract disagree
6 of the vaults checked publish a withdrawable amount their own contract contradicts. We show them rather than dropping them, because this is the figure the exit register rests on and an overstated one makes stuck capital look free.
Where the age is shown, the two network routes to this figure,cash() and totalAssets() − totalBorrows(), agree with each other, so the gap is how old our reading is rather than how we compute it. That makes it a refresh rate, which does not make it harmless: the register still overstates the exit until we read again.
| Vault | We publish | The contract says | Ours ÷ theirs | Our reading |
|---|---|---|---|---|
| euler-usdc-797dd8 | $187.0k | $159.8k | 1.17× | 0.8h old |
| euler-usdc-4c1aed | $133.7k | $144.2k | 0.93× | 0.9h old |
| euler-usdc-6afb8d | $99.2k | $96.0k | 1.03× | 0.9h old |
| euler-aero-5fe2de | $81.0k | $72.5k |
Does the money add up?
4,679 daily windows across 201 vaultsA vault’s value moved by some amount. Money arriving explains part of it, closed trades another, fees, funding and open positions the rest. Whatever the five do not account for is published as unexplained rather than folded away, and this is how often that leftover is large.
| Unexplained, as a share of what moved | Windows | Share of them | Unexplained |
|---|---|---|---|
| under 0.1% | 3,315 | 70.8% | $11.8k |
| 0.1 to 1% | 942 | 20.1% | $30.5k |
| 1 to 5% | 266 | 5.7% | $305.2k |
| over 5% | 156 | 3.3% | $1.03M |
The largest leftovers, and what they have in common. Every one of them is a window in which money left the vault. A window where money arrived, at any size, closes to a fraction of a per cent.
- 2026-09-27drkmttr$181.7k unexplained on -$1.02M of flow
- 2026-09-22Jade Lotus Capital$170.3k unexplained on -$5.4k of flow
And the payments, checked the same way
read 2026-10-1050 payments re-read from the transactions they came from, all of them agreeing. An x402 row is not one event: it is an authorisation matched to a transfer, and the pairing is what everything published about payers, payees and amounts rests on.
A sample rather than a sweep. Four million payments cannot be re-fetched, and the matching is one rule applied identically, so it is right for all of them or wrong for all of them. 10 receipts could not be read and are counted apart from agreement.
What we can see being used
Knowing how much a vault holds and knowing what it is doing with it are different readings, and this site has far more of the first than the second. The gap is ours, not a finding about the money.
Finding, watchThe blind spot, measured
$11.29B
tracked across 2,586 vaults
$730.97M
whose deployment we can read
$10.55B
we can only see the size of
We can read the deployment of 6% of the capital we track. For the rest we hold a value the venue reports and no reading of what it is invested in, which is a limit of what a network publishes rather than a fact about the strategies.
How this is counted
Three kinds of vault need three different readings, and one number over all of them would be wrong three ways. A perp vault deploys into positions; a lending vault deploys by being borrowed from, so utilisation is the reading; a strategy vault reports a value and nothing else, and no venue reports what it does with it.
And what this still cannot say. Where we do read positions, a hedge placed on a centralised venue is invisible to us: a book that looks one-sided here may be flat, and the operator would look reckless for being prudent. An attestation would close that gap with a claim, and a claim is not a measurement. So the gap is published instead.
The network, end to end
Four steps. A strategy can enter at the second one without anybody asking it to, which is the point: the record exists whether or not the operator participates, and participating is how it gets stronger.
- 1
Declared
0signed declarationsThe operator signs what a strategy is, what it trades, what decides, the limits, from the address it trades from. Nothing before that signature is attested, and every passport says how many days ran first.
None yet. 3 demonstrations exist so the format is visible; counting them as adoption is the failure this site is built to refuse.
Look at it → - 2
Observed
1,284records checkable at sourceWe rebuild the record from on-chain execution and venue history, for every strategy in the perimeter, asked or not. This is what a grade of B means and it is not optional.
2,588 more are reported to us and could not be checked directly.
Look at it → - 3
Anchored
1batchA day of declarations and commitments reduces to one Merkle root, published on chain. After that the database cannot be quietly edited: the root would stop matching.
2 anchors on chain; 2 decision commitments, 2 revealed.
Look at it → - 4
Graded
Registry
Every strategy anyone declared, including the ones that went wrong.
A declaration is never deleted and never edited, and the outcome column is filled from what we observe rather than from anything the operator tells us later. Its value is entirely in the rows nobody would submit after the fact.
Certification
What each grade requires, and what none of them mean.
Three rungs, rated from public data, pre-registered before it ran, every decision committed before execution. A grade never raises a rating, and a strategy that lost money with a complete record outranks one that made money on a claim nobody can check.
Proofs
The material to check any of this yourself.
The exact text that was signed, the signature, the digest, the batch root and the inclusion proof, with the steps to recompute each one.
Capital anybody can look up, and capital nobody can
A vault reaches us either with its own contract address or with an aggregator’s pool identifier standing in for one. The second kind cannot be checked: no governance reading, no share price of our own, no independent confirmation of a single figure we print about it. We are repeating what one source said, which is a defensible thing to do and an indefensible thing to do quietly.
Published capital
$11.29B
across 2,586 open vaults we can size.
Rests on one source’s word
$83.58M
1% of it, over 46 vaults with no on-chain identifier we can read.
Judged per network
16
venue-and-network pairs affected. A Solana account is not an EVM address, and testing for the wrong shape reported our own pattern as their missing data.
| Venue | Network | Vaults | Capital | Of the unconfirmable |
|---|---|---|---|---|
| Ember Protocol | sui | 11 | $38.92M | 47% |
| Fusion by IPOR | base | 2 | $15.49M | 19% |
| Neutral Trade | solana | 10 |
How much of the book we have read for manager activity
Whether a vault’s contract moves assets on its own account is read directly from the chain, one batch of vaults per run, because the largest networks cannot be swept inside a single run’s budget. A vault not yet reached carries no reading at all rather than a quiet one, below is how far that has got.
| Network | Read | Vaults with an address | Share | Oldest reading |
|---|---|---|---|---|
| BSC | 0 | 840 | 0% | — |
| Robinhood Chain | 315 | 672 | 47% | 4 September 2026 |
| Base | 185 | 565 | 33% | 21 September 2026 |
| Ethereum | 438 | 461 | 95% | 27 September 2026 |
| 1 | 324 | 0.3% | 4 September 2026 | |
| Arbitrum | 118 | 121 | 98% | 6 October 2026 |
How old the figures are
Every figure on this site carries the moment it was read, and nowhere did the site say how that is going in aggregate. Below is each venue’s normal state and its worst case, because one venue refreshing a thousand vaults hourly with a single straggler is not the same as a venue that has quietly stopped.
Re-read within a day and a half
95%
of 2,585 open vaults that report capital.
Older than that
126
vaults holding $3.56M, <0.1% of the capital here.
The threshold
36h
Long enough that an hourly reader missing a run is not an incident; short enough that a venue which has stopped shows up the next morning.
| Venue | Vaults | Capital | Typical age | Oldest | Over 36h |
|---|---|---|---|---|---|
| 222 | $5.08B | 1h | 5h | — | |
| 163 | $2.37B |
What this section cannot establish
Nothing here proves an operator has no second, unregistered book. That defence is contractual and economic rather than cryptographic, the covenant is binding, declared addresses are cross-checked against inferred siblings, and an A grade has to be worth more than the deception. It is stated here rather than left for someone else to discover. The methodology page defines every count above.