Evaluating a DeFi vault means understanding what it is built from. Beneath a single deposit interface sits a stack of distinct components: a token standard that handles share accounting, a vault type that sets the scope of what the vault can do, a curation layer where a manager defines the risk parameters, and a set of controls that limit what that manager is permitted to change.
This article defines each component and the vocabulary that goes with it, so that any specific vault can be read rather than trusted. It stays neutral on which implementations are best.
For context, a vault is a smart contract that pools deposits and deploys them into yield-generating strategies, issuing depositors a token that represents their proportional share. The questions that follow are what that contract is allowed to do, who decides, and how a depositor can verify it.
The Share-Accounting Standard
Most vaults on Ethereum mainnet are built on ERC-4626, a tokenized vault standard that creates a uniform interface for how deposits, withdrawals, and share accounting work.
The mechanics are direct: a user deposits one asset and receives a fungible share token, sometimes called a receipt token, that represents a proportional claim on the vault's total assets. The exchange rate between the asset and the share token is defined by the vault and changes over time as yield accrues. When the user exits, they redeem share tokens for the original deposit plus the returns that have accumulated since deposit. The vault therefore tracks both each user's principal and their proportional share of all returns generated, without needing to record individual positions.
Before the standard existed, every vault protocol implemented its own deposit and withdrawal logic, which made it difficult for other protocols to build on top of vault positions. A shared interface changed that. Vault share tokens can serve as collateral in lending markets, trade on exchanges, or compose into higher-order products. This composability is one of the key enablers of the products-built-on-vaults trend, and it is the foundation for the structured products that now sit on top of vault infrastructure. Standardisation is what turned the vault from an isolated product into a building block.
The exchange-rate mechanic is worth stating precisely, because it is a frequent source of confusion. A vault share does not increase in number as yield accrues. The number of shares a depositor holds stays constant, while the amount of underlying asset each share can be redeemed for rises. A vault that has generated returns will redeem each share for more of the underlying asset than it cost at deposit. This is why two vaults can quote very different share prices while offering similar forward returns: the share price reflects accumulated yield since inception. An evaluator reading a share price is reading history, and the rate that matters for a new deposit has to be assessed separately.
Non-Custodial Design
A defining property of a well-designed vault is its non-custodial nature. Depositors retain ownership of their assets and can withdraw subject only to available liquidity and any withdrawal-speed restrictions. No intermediary can redirect funds.
The property is enforced at the smart contract level, which makes the custody guarantee as strong as the underlying code. The degree of assurance varies across vault types. Simple vaults offer instant exit when liquidity is present. More complex vaults may introduce cooldowns or staged withdrawals so positions can be unwound cleanly. In both cases the funds can only be returned to the depositor, so the property holds while the time to liquidity differs.
This is the first point in the stack where the two main vault types diverge in a way that matters to a depositor planning an exit. The simpler type behaves like a lending position, where the constraint on withdrawal is whether other depositors have left enough liquidity in the market. The more complex type behaves more like a fund with a redemption process, where exiting may require the orderly unwinding of positions held across several venues. A depositor who needs predictable access to capital should treat the withdrawal mechanics as a primary design feature rather than a detail to check later.
Two Categories of Vaults
There are two primary categories, defined precisely here.
An earn vault is a single-protocol vault. A user deposits one asset, commonly a stablecoin, and the vault distributes it across isolated markets within one lending protocol. Each market has its own collateral type, loan-to-value ratio, and interest rate. Withdrawal works the way it does in any lending market: if liquidity is available, the user exits. Earn vaults are the simplest and most widely adopted modern design, and they carry the strongest non-custodial guarantees because the manager's powers are tightly constrained by the protocol.
A multi-strategy vault operates across protocols, and in some cases across chains. It accepts one asset and deploys it into several strategies at once, rebalancing across them based on yield, risk, and liquidity conditions. Its scope and flexibility exceed an earn vault's, and so does its complexity. A single multi-strategy vault might run a leverage loop, provide liquidity on an automated market maker, and hold a basis trade at the same time. Withdrawals may involve cooldown periods, because unwinding positions cleanly takes time.
The two categories share a label in common usage and differ in mechanics and risk. A reader evaluating a product should look past the label to the architecture, the strategy composition, and the risk controls.
The distinction is critical for evaluation because the two categories fail in different ways and demand different diligence:
An earn vault's risk is largely legible from its parameters, because the manager's discretion is bounded by what the protocol permits.
A multi-strategy vault's risk depends on the manager's ongoing decisions, which cannot be fully read from a parameter set captured at a single moment.
A reader who applies earn-vault diligence to a multi-strategy vault will understand only part of the risk they are taking, because the part that matters most is the part that changes over time.
The Curation Layer
An earn vault has two layers. The base layer is the lending protocol, which holds the individual markets and their pairings of supply and collateral assets. The vault layer sits on top and is designed by an entity called a curator.
The curator does not custody funds or move them at will. Instead, it sets the parameters that govern how deposits flow into the markets below, and the contract enforces those parameters.
The core curator parameters are the following:
Collateral selection: Which assets a market accepts. This decision defines the vault's risk profile more than any other parameter, because the quality and liquidity of the collateral determines what happens when a position must be liquidated. It is set at the contract level so that only vetted collateral can be used by the vault.
Maximum loan-to-value ratios: The amount a borrower can draw against collateral. Higher ratios increase capital efficiency and increase the risk that a price move pushes a position into bad debt.
Liquidation thresholds: The point at which a position becomes liquidatable, which may be set separately from the borrowing limit to create a buffer between borrowing and liquidation.
Supply and borrow caps: Limits on how much of the vault's capital can be exposed to any single collateral type, which contain the damage from a single asset failing.
Interest rate models: The cost of borrowing at different utilization levels, which drives both the yield the vault generates and the liquidity available for withdrawals.
Oracle configuration: The price feed that determines collateral value for all liquidation calculations. On some protocols this choice is fixed once a market is created, so it cannot be corrected later without migrating to a new market.
For a user lending a stablecoin, this is an accessible entry point. The user deposits into the vault, and the vault allocates the funds across the markets within the parameters the curator set. The parameters are usually transparent and accessible to everyone, and changes to them are usually time-locked, which is what lets the depositor verify the vault rather than trust it.
These parameters interact rather than operate in isolation. A generous loan-to-value ratio paired with a thinly traded collateral and an aggressive interest rate model produces a different risk than any one of those settings would suggest on its own. This is why the curator's role is closer to portfolio construction than to filling in a form. The parameters define a system, and the behaviour of that system under stress is the thing actually being designed, whether or not the curator frames it that way. Evaluating a vault therefore means evaluating the combination of parameters, not each parameter in turn.
In a multi-strategy vault, the equivalent design role is the strategist, who decides which strategies the vault deploys into and how to weight them. The strategist holds wider discretion than a curator, because the surface is larger: multiple protocols, multiple position types, and sometimes multiple chains. Depending on the contract design, many of these vaults allow strategies to be added, removed, or reweighted as market conditions change, within preset structures. The wider the discretion, the more the depositor depends on the strategist's judgment and process rather than on a fixed set of constrained parameters.
The wider discretion changes what a depositor is evaluating. With an earn vault, the depositor can evaluate a largely fixed configuration. With a multi-strategy vault, the depositor is evaluating a decision-making process and the people or systems running it, because the configuration will change over time. The questions shift from what is the current allocation to how allocation decisions are made, what constraints bind them, and how quickly the vault can exit a strategy that stops working. These are questions about governance and process, and they cannot be answered by a single snapshot of the vault's holdings.
Separation of Powers
The role architecture within a curated vault is typically multi-layered. A platform provider supplies the underlying protocol, an infrastructure and accounting provider handles share accounting and reporting, and a strategy provider designs the allocation. Separating these functions means no single party controls the whole system, which limits the damage any one of them can do.
Two further mechanisms constrain curation power directly. A guardian, sometimes called a sentinel, monitors markets in real time and can pause operations if an exploit or anomalous activity is detected. This role is often a third-party security service, such as Hypernative. The guardian holds no allocation or parameter-changing power. It can only freeze the system as a protective measure. This separation provides a check against external attacks and against curator error, because the party that can stop the system is not the party that runs it.
A timelock governs parameter changes. Because curators have the authority to change parameters, most protocols require that significant changes pass through a delay, typically several days, before they take effect. This gives depositors advance notice and the ability to withdraw beforehand if they disagree with a change. Combined with emergency pause functionality, the timelock balances the curator's need for operational flexibility against the depositor's need for predictability, transparency, and an exit.
The separation becomes relevant most precisely when something goes wrong. A guardian that can freeze but not allocate cannot be used to extract value, only to protect it, which is why the freeze power can safely sit with a third party. A timelock that delays changes gives depositors the one thing an opaque product never offers, which is time to react before a change takes effect. Neither mechanism prevents a curator from making a poor decision. Both ensure that a poor decision becomes visible and that depositors have a path out before it binds them. The design assumes curators are fallible and builds the depositor's protection around that assumption.
Reading the Stack as a Whole
The stack reads cleanly from bottom to top. The lending protocol holds isolated markets. The vault layer, designed by a curator or strategist, allocates across those markets within set parameters. A guardian watches for failure and can freeze the system. A timelock slows changes so depositors can react. Share tokens issued by the vault can compose into products built one level higher.
Every layer is a place where a decision is made and a risk is taken. Reading a vault means locating where each risk sits rather than treating the deposit as a single, undifferentiated product. A depositor who can name the token standard, the vault type, the curator's parameters, the guardian, and the timelock of a given vault knows what they are holding. A depositor who cannot name them is trusting the label alone, which this market has repeatedly shown to be unreliable.
About this article: This article is based on the data and analysis in Sentora's research report, The Vault Economy: Architecture, Risk, and the Rise of Professional Curation in DeFi. Read the full report for the complete figures, sources, and detail.
