Loading
How good is this data today
Which readers ran and when, how much of what they set out to read they reached, where our figures disagree with somebody else’s, and what we refuse to publish rather than estimate.
How this is read
Everything here was already somewhere on this site; no page put it together.
There is no single quality score, deliberately. One number would average a stale venue against a refused price against an unexplained gap, and those three fail differently and matter to different readers. The components are here; the composite is not.
225
Readers on a schedule
12 more are one-off tools, which cannot be late
3
Late
past twice its own interval, listed below
10
Partial reads
a total from these is a floor
1
Readers failing
a reader did not complete its last run
2
Assertions failing
independent sources that disagree, which a healthy reader cannot catch
What is claimed and what is paid for do not meet
Three measurements taken separately, for different reasons, turned out to be the same finding: the parties who declare something and the parties who get paid are, in our reading, nearly disjoint sets.
Why each row names its populations
Each row carries the exact populations it compares, because the same question asked with a looser definition gives a different number and both are true.
| Two populations | One side | The other | In both | Measured over |
|---|---|---|---|---|
| Paid on a rail, and a vault we read | 3,748 | 3,926 | 0 | Addresses ever paid for a SALE (the recipient lists a price in a catalogue we read) against open vaults, matched on the vault’s contract address. |
| Holds an ERC-8004 registration, and runs a book | 1,449,067 | 481 | 0 | Distinct registration owners against operators carrying an on-chain address, matched case-insensitively. |
| Declares a callable endpoint, and has been paid | 3,090 | 3,748 | 15 | Owners declaring any service kind (MCP, A2A, OASF, web, wallet, social, x402, other) against the paid addresses above. /work asks a looser version of this question (MCP and A2A only, any incoming transfer counted as paid) and answers 15 of 1,287; both are true. |
This is what we can see, not what exists. One firm can run both sides under addresses that never touch, exactly as one owner can fund payers that never pay each other. No link is visible; that is not the same as no link.
What each verdict rests on
in-line, beating and short are the sharpest thing said here about a vault: what a holder actually earned against what the venue advertises. A verdict drawn from a price that moved once is arithmetically right and describes nothing that will happen again, so this is how many of them are which. Counted over the 424 open vaults whose realised return can be compared with an advertised one at all.
| Verdict | Vaults | Movement spread | One step explains it | Series barely moved |
|---|---|---|---|---|
| in-line | 369 | 365 | 2 | 2 |
| short | 29 | 25 | 3 | 1 |
| beating | 26 | 25 | 1 | 0 |
The two weak columns are never added: only the first is a thing we can name.
What the two columns mean
One step explains it: enough intervals moved to judge, and one of them still carries most of the window: a correction, or a fee taken in one go. Series barely moved: most of the movement sits in one interval of a price that hardly moved at all, which is a thin reading rather than a step we can account for. The two are never added: only the first is a thing we can name. Every vault says which of these applies to it in the field of its own check.
What a realised return here is made of
A realised return missing a component is not a smaller return. It is a different measurement wearing the same name, and which way it is wrong depends on which component is missing. Counted over the 2,621 open vaults holding capital, $12.25B between them.
| Component | Vaults | Share | Capital behind it | What it means where it is missing |
|---|---|---|---|---|
| A value series at all | 2,618 | 100% | $12.25B | Without one there is no realised return to state, and the vault is absent from every return figure rather than counted as flat. |
| Profit stated by the venue | 1,682 | 64% | $241.31M | The venue’s own profit series. MoreWhere it is missing the return is reconstructed from value alone, which cannot separate a deposit from a gain. |
| Share price | 871 | 33% | $11.88B | What one share was worth. MoreIt is the only series that survives deposits and withdrawals untouched, so a vault without it needs flows read correctly instead. |
Readings we had and no longer have
8 open vaults lost a reading since the previous check, and EVERY ONE of them says why. The largest is a capital figure (tvl), lost by 8: withheld: the venue has not reported this vault since 2026-10-29, while it reported others on the same venue in the last two days. A reading withheld on purpose is this working rather than failing; it is listed here because a withheld figure must not be quiet either.
- a capital figure8 vaults · tvl
Krystal · Commo ++, Krystal · Sky, Krystal · KFB BAAK, Krystal · Vandi v2, Krystal · Nebeng Paman yang BSC, Krystal · FomoBSC, and 2 more
Every one says why: withheld: the venue has not reported this vault since 2026-10-29, while it reported others on the same venue in the last two days
Compared against the previous check on 2026-11-05, over 3,926 open vaults. A vault that CLOSED takes its readings with it and is not counted. This says what changed between two checks, not how long something has been missing.
How old is the capital figure, venue by venue
A size read two days ago is not wrong; it is old, and the difference matters to somebody sizing an exit. A venue is called stale here after 36 hours.
| Venue | Vaults | Median age | Stale | Capital behind them |
|---|---|---|---|---|
| 237 | 1.0h | 0 | — | |
| 165 | 1.0h | 0 | — | |
| 31 | 1.0h | 0 | — | |
| 26 |
Readers that are behind
Every figure is as old as the reader behind it.
How this is read
These are past twice their own interval, or a quarter of an hour where that is longer, so anything resting on them is older than it should be. A reader that runs weekly and completed two days ago is not here, which is the point of measuring against its own schedule rather than against a flat day.
- vault-flagslast run failed 4 hours ago · runs every 720m
- activitylast completed 10 days ago · runs every 15m
- tvllast completed 10 days ago · runs every 240m
- depositorslast completed 10 days ago · runs every 360m
How we check ourselves
Four questions asked of this site every six hours, each with an answer that does not share an input with the figure it checks: the published size of the largest vaults against the contract’s own quantity at an outside price, identity registries against their own id sequence, readers that succeed and get nowhere, and registers whose published count has not moved while vaults wait.
How this is read
A failing answer reaches the operator the same way a stuck reader does.
Last asked 126 minutes ago.
- sanity check passedPublished capital against the chain and an outside price
59 of 59 largest vaults pass the ±25% sanity check, 0 fail it; tighter: 56 within 1%, 57 within 5%, 1 could not be asked
What this check cannot reach
- a sanity check at ±25%, not a reconciliation: our figure is as of our last refresh of each vault, the chain's quantity is read at this run's latest block, and the price is the outside feed's current quote, so the two sides can be hours apart
- askable this way: 817 vaults, $11,122,646,781; the 60 largest of them are read every run
- not ERC-4626: no standard call answers what the vault holds: 1,389 vaults, $757,054,693, not checkable this way
- Hyperliquid: an order book; the vault is a margin account, no contract holds the capital: 336 vaults, $254,534,502, not checkable this way
- Solana: the venue publishes the size; there is no contract that answers totalAssets(): 52 vaults, $79,188,169, not checkable this way
- no contract address on record: 27 vaults, $40,442,336, not checkable this way
- agreesIdentity registries against their own id sequence
20 chains, 1,710,454 identities held, 0 chains with a hole larger than 1%
How well we know whether declared endpoints answer
2,637 distinct hosts are declared by registered identities, and the last probe reached 2,352 of them, 89.2%.
How this is read
285 were not reached before that run's deadline expired, so their state is what an earlier day found rather than what is true now. A host we did not ask is not a host that is fine. That probe ran 0.0 hours ago. 2,159 of those hosts are in a state that is the endpoint's own fault and 3,396 identities stand behind them. The rest are not one group: a 402 is a seller asking to be paid and working, a 401 is credentials we do not hold, and a timeout describes our attempt rather than the host. Separately, and on its own denominator, 103,722 x402 resources are listed by sellers. We have knocked on 41,969 of them at all, 40.5%, and only 32,341, 31.2%, have been knocked the 3 times it takes before we will publish a state, because one failure is not a finding. So the honest reading of the second population is that we can say almost nothing about it yet, and that is our reach rather than a judgement of the sellers.
| What the host said | Hosts | Share | Identities behind | Is it the endpoint failing |
|---|---|---|---|---|
| AnsweringA 2xx. The endpoint is there and served us. | 356 | 13.5% | 12,405 | no, it is working |
| Asking to be paidA 402. An x402 seller doing precisely what it advertises, which is the opposite of a fault. Whether the offer can be signed is a separate reading: a HEAD request never sees the body an offer lives in, so that one comes from the knock flow and has its own vocabulary of eight faults. | 7 | 0.3% |
Coverage, measure by measure
How much of each thing we can actually see.
How this is read
Five separate questions with five separate populations, so the percentages are not comparable with one another and are never added up. Where capital is missing, the largest venues it sits on are named.
| What | Share | Seen, of | Largest gaps |
|---|---|---|---|
| Perp-vault capital whose open positions we read in the last three daysPopulation: open perpetual-venue vaults with a size. Capital, not notional: a vault counts by its equity, however levered its book. | 38.4% | $92.42M of $240.68M | Hyperliquid $146.32M (65) · dYdX $1.93M (0) |
| Position notional in a market whose open interest we hold, so a crowd can be sized against itPopulation: gross notional of the positions we read, per instrument. Notional over notional. It is never divided by vault capital, which would let a levered book inflate the figure. | 2.1% | $170.80M of $8.27B | venues that publish no open interest |
| Lending capital whose collateral we can see throughPopulation: open lending vaults with a size. Includes vaults that lend nothing, read as such from their own totalBorrows(). | 99.8% | $9.39B of $9.41B | Euler $19.43M (28) · Morpho Vaults V2 $1.51M (3) |
Where we disagree with a public source
Several figures here are also published by somebody else, and several of them differ.
How this is read
Each comparison carries an explanation or is marked as not having one.
13
explained
1
investigating
What we refuse to publish
These counts are a measure of honesty, not of coverage. A vault whose underlying we cannot price is published as unvalued rather than estimated. Reading the number below as a failure is the one misreading this page exists to prevent: the alternative is not a better figure, it is a made-up one.
1,291
Open vaults with no size
of 3,926 open
no price we trust for the underlying
2,724
Open vaults with no underlying read
of 3,926 open
the contract does not tell us what it holds
Methodology in force
Figures and evidence aspern-0.5 · assessment document aspern-0.4
Every published figure carries the first in its evidence block.
How this is read
The hashed assessment document carries the second, which moves for its own reasons, and a document can be rebuilt at an older version so a hash issued under it stays checkable against the reading of its own day.
How each figure is arrived at · what changed and when · the independent check