Trust and safety
Three questions worth asking about a company that rates other people's vaults: what does it hold, what can still go wrong, and can its ratings be checked. Here are the answers, including the two times we cost users money.
1. We never hold your money
AtlasYield deploys no smart contracts. There is no AtlasYield pool, no vault of ours, no admin key over your assets, and no contract of ours on the path your funds take. When you deposit, the transaction goes from your wallet to the protocol, and you sign it yourself.
This is checkable rather than asserted. Open any deposit in a block explorer and the counterparty is the protocol, not us. We also take no fee on deposits and no share of yield, which is a deliberate constraint and not an oversight: revenue that scales with deposits would pay us never to say do not deposit, and that is the one thing a rating has to be able to say.
2. What can still go wrong
Not holding funds is not the same as being harmless. The app builds transactions that you sign, and a badly built transaction can lose money even though we never touched it. That has happened twice. Both are on-chain and permanent, so here they are.
Shares credited to the router, not the depositor
BaseConfirmed loss of user funds
Some protocols mint vault shares to whoever calls the deposit function. When a deposit was routed through an aggregator, the aggregator became the caller, so the shares were credited to the router contract and the depositor received nothing. The transaction succeeded on-chain, which is what made it dangerous: nothing looked wrong.
What prevents it now: Every deposit path is checked against a list of protocols known to mint to the caller, and any unsafe routed path is refused outright and forced to a direct deposit. This is a fail-closed invariant with its own test suite, and a new deposit surface that skips it is treated as a release blocker.
A route that succeeded and returned dust
Ethereum, 13 August 2026$10.00 deposited, $0.00077 received
A Pendle market advertised roughly $1M of liquidity but had no tradable depth. A $10 deposit returned a position worth less than a tenth of a cent. Both existing safeguards passed honestly: the shares did reach the user, and the balance did change, so the deposit was recorded as confirmed. The failure was that nobody was checking whether the value that came back resembled the value that went in.
What prevents it now: Every deposit path now asserts that at least 85% of the input value is returned before any approval or send, and it fails closed when the output cannot be priced. Separately, a daily screen probes a full round trip on each vault using the path that protocol actually executes, and vaults with no working exit are removed from the catalog rather than offered.
One scoping note, stated plainly because it matters for anyone evaluating the data product rather than the app: both incidents were in the consumer deposit path. The ratings API touches no funds, moves no money, and has no transaction surface at all.
3. Is the rating itself audited?
No. No third party has reviewed the scoring engine, and we are not going to imply otherwise with a badge. Because we deploy no contracts, a smart-contract audit of the usual kind would have nothing to examine, and a code review of the engine would attest to one version on one date.
What we offer instead is stronger for a ratings product, and harder to fake: every score we have ever published is committed to a public git repository with a timestamp. Anyone can check what we said about a vault on a given day, against the commit that said it. We cannot quietly revise a bad call, because the old one is still there.
- Scoring has run since 2026-05-05. Public, verifiable snapshots begin 2026-07-08.
- 503 vaults carry a published score as of 2026-08-30, and each one has its own page stating when it was measured.
- 43 of those scores are built on incomplete inputs. They are labelled as such on the index and on each vault page rather than being hidden or quietly dropped.
- Scores are produced automatically from public data. They can be wrong, stale or incomplete, and they are research and information, not investment advice.
Read the published history · How the score works · Calls we have made
4. Who runs AtlasYield
AtlasYield is built and operated by Jeevesh, working solo. The engine, the API and this site are one person's work, which is worth knowing when you weigh a rating that comes from it.
Reachable at admin@atlasyield.club, on X, and on GitHub, where the score history is published. If something here is wrong, tell us and we will correct it on this page.
5. What we do not take money for
No protocol pays us to be rated, to be rated higher, or to be included. We take no fee on deposits and no share of yield. We have declined revenue-share arrangements on that basis. Terms for reusing our ratings are on the data licence page: free to cite with attribution, paid only for volume and guarantees.
This page is maintained as incidents occur. It may never omit one. Canonical version: https://atlasyield.club/trust. Atlas scores are research and information, not investment advice. DeFi carries risk of total loss.