A Zero Drawdown Is an Accounting Choice
Drawdown cannot move until someone books a loss. Sharpe divides by that same decision, age measures a deployment date, and fee has been competed down to 0.205% of assets. Four queries that would actually separate one curator from another cost one log call and a handful of getters, and nobody runs them.
Every column an allocator sorts curators by is an output of the curator's own choices.
A vault's share price cannot fall until someone decides to book the loss, and whether anyone does turns on which version of the vault contract was deployed. Sharpe divides by that same decision. Age measures a deployment date. Fee has been competed to 0.205% of assets, a spread too thin to rank anyone on.
The facts that would separate one curator from another are not reported by the curator. Thirty-two mainnet V2 vaults have a curator that is a single key or a Safe with a threshold of one, and they hold $1,065.2M, 50.6% of mainnet V2 assets. No ranking prints it.
Same loss, one drawdown
Two vaults can take the same loss, in the same market, on the same day, and only one of them prints it.
Which one a depositor lands in was settled before they arrived, by whoever chose the contract version. A Vault V1.0 "automatically realizes the loss on all existing depositors at the moment of the bad-debt liquidation event", and its share price "will instantaneously decrease." A Vault V1.1 does "not automatically realize the loss". "The share price will not decrease." The debt stays in the market and "The last depositors will face the bad debt unless action is taken." All five quotes from docs.morpho.org/curate/tutorials-v1/bad-debt, read 2026-08-25.
Sort the whole market on drawdown and you have sorted three vaults. Of the 485 vaults in our index above a $100,000 floor, 244 publish a computed maximum drawdown and 234 of those publish exactly zero, holding $32.00B between them, 97.4% of the assets in vaults that report the metric at all. Ten report anything else, and seven of the ten are tokenized shares in offchain funds. The three that hold onchain positions read 3.87%, 1.96% and 0.89%.
Publishing a drawdown at all is optional, and 241 of the 485 decline. 227 publish no inception date either, so both shares are computed on the half that answers. Distrust our census before you distrust the argument: it is our own index at priime.finance/ecosystem, built from our reads of DefiLlama, Morpho, Beefy, Yearn, Euler, IPOR and Lagoon plus direct on-chain calls, snapshot 2026-08-25 09:20 UTC, refreshed daily. An empty ledger is not a clean one.
Morpho's two largest mainnet USDC vaults fund one market each, and it is the same one. Adpend USDC holds $183.6M, enables 30 markets and funds one. 1337 USDC holds $97.9M, enables four and funds one. The market is 0x0f9563442d64ab3bd3bcb27058db0b0d4046a4c46f0acd811dacae9551d2b129, USDC lent against sdeUSD, which reports utilization of 1.0000, withdrawable liquidity of $0.00, and a supply rate of 297,996%. Morpho flags its oracle oracle_unusable at RED. On 2025-11-20 Adpend held $483,805 and one share was worth $1.05. On 2026-08-20 one share was worth $179.04. Over the same stretch 1337's share went from $1.08 to $361.14. Vault and market state from Morpho's GraphQL API, fetched 2026-08-25.
The assets column is not a record of deposits. Nobody deposited $183M into Adpend. The figure is compounding at the rate of a market that cannot pay out a dollar.
A vault that enables thirty markets and funds one is a single-market vault carrying a diversification story. Enabled is a cap the curator set. Funded is where the money actually sits, and the same API response returns both. The subtraction points at a vault. Then read the market.
The largest reported position in Morpho's mainnet USDC book is invisible in the benchmark we publish. Neither vault is in our index, because our sources drop unlisted vaults. That is our gap, and it is one query wide.
Sharpe inherits the denominator
The higher the Sharpe in this table, the less it is telling you. The 199 vaults publishing a Sharpe have a median of 15.09, and across the 189 publishing both a Sharpe and a zero drawdown the median is 15.45; 78 of them read above 20. Those are not risk-adjusted returns. They are what the formula prints when the denominator is a line that only goes up.
What the ratio was standing in for is a count the chain already keeps. A Morpho Vault V1 emits ReallocateSupply and ReallocateWithdraw every time an allocator moves depositor capital between markets (Ethereum mainnet, 2026-07-26 to 2026-08-25, joined to the 81 mainnet V1 vaults above $100,000; V2 allocates through adapters and emits a different set). Filter the two topics, count distinct transactions per emitting address.
Nothing an allocator sorts on distinguishes a vault touched 1,083 times in a month from one touched once. Steakhouse USDC logged those 1,083 transactions on $73.3M, 36 a day for the whole window. Smokehouse USDC recorded 1,160 on $14.7M. Vault Bridge WBTC recorded one, on $29.2M. Gauntlet USDC RWA recorded two, on $11.2M. Across the 47 listed mainnet vaults the median is 18 transactions in thirty days, and among the twenty listed above $5M it is 52. Thirty-four of the 81 recorded none at all, holding $333.4M, 41.7% of mainnet V1 assets above the floor.
A zero in that count can be the whole product. Hakutora USDC holds $17.5M, enables four markets, funds one, charges a 15% performance fee, and did not reallocate once in the window: one market, one rate, no discretion to audit. That is a different business from a vault turning over positions 36 times a day. The table is sorted on neither. Ask the zeros which one they are.
A track record needs a bad month
Age is the column that looks hardest to fake, and it measures a deployment date. Of the 258 vaults publishing an inception date, 126 are under a year old and hold $18.05B, 54.9% of the dated set. A vault's worst month is not in its history if the history is nine months long.
An old vault reporting zero is making a much larger claim than a young one. 129 of the vaults reporting exactly zero are at least a year old and hold $14.80B, and every one of them was open through November 2025, the month Morpho Blue went from $8.256B on 2025-11-01 to $5.808B a month later. Whatever those vaults lived through in that window, the column records none of it.
The zero that should worry an allocator most is the one attached to the oldest and largest vaults. In our index, 18 vaults hold $100M or more, have been open at least a year, and report a maximum drawdown of exactly zero. That is $13.33B allocated against a number that has never had a reason to move. The inception date sits beside the drawdown on the same page, so the check costs nothing. Ask those eighteen what November cost them.
Half the money is one signature
The fee argument is over and depositors won it. Take each live V1 vault's gross APY, subtract its net, multiply by assets, and sum: across the 138 vaults with a gross rate at or below 50%, holding $1.6688B between them, the difference runs to $3.42M a year, or 0.205% of assets annually for the entire selection function (Morpho GraphQL, 2026-08-25). What the winning price buys is a list of addresses.
Morpho already publishes who should hold each key, and its answer is not the same for every role. For the V2 owner: "A single EOA is highly discouraged for production vaults." For the curator: "A multisig (e.g., 2-of-4 or 3-of-5) or an MPC wallet." For the allocator, a hot key is explicitly endorsed: "A hot-key EOA operated by a bot, or a fast-response multisig." From docs.morpho.org/curate/concepts/roles, read 2026-08-25. A bot key moving capital is the standard working as designed.
The curator key is where the deployed set stops matching that recommendation, and it is where the money sits. At mainnet block 25,834,145 there are 111 live V2 vaults above $100,000 holding $2,104.1M. Thirty-two have a curator that is an EOA or a Safe with a threshold of one, holding $1,065.2M, 50.6% of mainnet V2 assets. Of that, $892.5M sits under a single Safe with one owner and a threshold of one, and that owner is an EOA: 0x9e396dE3312D373b87F9BD8763fb48184b42aac0. Twenty-two have an owner with no code deployed to it, holding $129.1M. On the V1 side, 11 of the 81 live mainnet vaults have an EOA owner, against documentation that reads "Never use single-key EOA (Externally Owned Account) control".
One transaction can change who is allowed to move that money. Adding an allocator carries a configured timelock of zero seconds on 92 of the 111, holding $1,873.6M. The role that can revoke a curator's pending action before it executes is the sentinel. Twenty-two of the 111 have none.
Reading the control surface needs no history, so it works on a vault that opened yesterday. Read the owner, curator and sentinel off each vault, check which of those addresses have code, and read the threshold where one resolves to a Safe. That is a handful of getters and one eth_getCode. Then ask the curator who the second signature is.
None of this is a disclosure failure. Morpho publishes the roles, caps, timelocks, fees and allocations through a free API, emits the reallocations on a public chain, and charges for none of it. DefiLlama serves nine months of daily protocol series without a login. The table is built from the other columns.
The same reading catches us. Priime's own operator set at launch is small and founder-run, and its upgrade authority is named rather than timelocked. Both facts are published. Run it on us.
Repricing takes nine months
The market does correct this, and the instrument it uses is a slow bank run. DefiLlama's Risk Curators category counted 41 protocols on 2025-11-01 holding $10.01B. On 2026-08-24 those same 41 held $8.51B, 31 of them smaller than they were and 11 now reporting under $100,000. MEV Capital went from $1,466.9M to $12.0M over that window and Sentora went from $251.0M to $2,416.3M. Daily series from api.llama.fi/protocol/<slug>, fetched 2026-08-25.
Nine months of withdrawals is not a diligence process you can run before you deposit. The money doing the repricing is already inside. It works. It works afterwards.
Price the decision
A position is public the moment it settles, and the reasoning that produced it is gone by the next block. The inputs an allocator read at the moment of the call are not stored anywhere an outsider can pin them. Every one of those queries is a proxy. Better than the column it replaces, and still a proxy for a thing nobody records.
A decision becomes checkable when two things hold: the inputs are pinned somewhere anyone can fetch them, and a party with no stake in the answer re-derives the same output and signs it. Neither is exotic. A Priime vault is built to settle a rebalance only when both do.
The queries were never the hard part. No allocator has made them the price of the money.
Put the stack to work.
Compose a vault from verified modules: delta-neutral LP, funding-rate carry, leveraged loops. Self-custodial, exit any time.