Why it matters
Every market needs a size, and onchain markets settled on this one early because it can be computed from public state without asking anyone's permission. Total value locked is the dollar value of the assets a set of contracts can be shown to hold. It appears in protocol dashboards, index construction, allocator screens and central bank financial stability reviews, and it is routinely the first number quoted about a protocol.
It is also the number most likely to mislead an allocator who reads it as a balance sheet. Two problems sit inside it. The first is arithmetic: a single deposit can appear in the totals of several protocols at once, because the receipt token issued by one is accepted as a deposit by the next. The second is denomination: the measure is a quantity of tokens multiplied by a price, so it moves with the market even when no participant has added or withdrawn anything. Neither problem is hidden, and neither is corrected in the headline figure.
How it works
A data provider maintains an adapter for each protocol, which lists the contract addresses and token balances that count, queries them at intervals and converts the balances to dollars at a reference price. DefiLlama, the most widely cited provider, showed total DeFi value locked of $75.833bn when accessed on 25 July 2026, led by Lido at $17.42bn, Aave at $14.254bn across 23 chains, Morpho at $7.497bn and Sky at $6.198bn.
The adapter is where the judgement sits. A BIS working paper published on 14 May 2025 examined how the figure is produced across 939 protocols on Ethereum and found that while 78.6% of them are read by standard balance queries, 68 alternative methods were in use, and 10.5% of protocols depend on external servers, so the reported figure cannot be independently reproduced from chain state alone. When the authors recomputed the number using onchain data only, their verifiable measure matched the published figure for 46.5% of the 400 protocols in their case study. They also found 240 identical balance queries appearing in more than one protocol's adapter.
Economic mechanism
Nobody pays for total value locked directly, which is what makes it interesting: it is a metric with no fee attached and therefore no discipline on its construction. What it does have is a price. Deposits attract fees, fees support token valuations, and token valuations fund incentive programmes that attract deposits. The figure is an input to the very flows it purports to measure, so it is reflexive rather than descriptive.
That reflexivity has a specific mechanical form. Liquid staking, restaking and lending each issue a claim token against a deposit. If a user stakes ether for a staking receipt, lends the receipt for a deposit token and posts that token as collateral elsewhere, three protocols report a deposit and the underlying asset is one. Luo, Feng, Xu and Tasca measured the gap between total value locked and what they call total value redeemable, the amount that could actually be withdrawn from the system once derivative and borrowed tokens are stripped out. At the peak of the 2021 cycle the two diverged by $139.87bn, on 2 December 2021, a ratio of roughly two to one.
The same layering makes the measure procyclical in a way a balance sheet is not. Because derivative tokens are collateral, a price fall triggers liquidations that unwind several layers at once. The authors found that a 25% fall in the ether price produced roughly $1bn more decline in total value locked than in the redeemable measure across six representative protocols, an excess that is bookkeeping unwinding rather than capital leaving.
Who bears the cost of the mismeasurement is worth naming. Depositors chasing an incentive programme are the marginal capital; the protocol treasury pays them in tokens; and the allocator who sized a position off a headline figure carries the difference between reported and redeemable when the layers unwind.
Participants
Data providers, whether DefiLlama or vendors selling institutional feeds, write and maintain the adapters, usually with contributions from the protocols being measured. Protocol governance bodies decide which contracts and incentive programmes exist and therefore what the adapter counts. Liquid staking and restaking operators create the receipt tokens that generate the layering; Lido reported 9,337,523 ether staked and $17.6bn of value on its own front page on 25 July 2026. Index providers, custodians and asset managers consume the figure. Central banks and supervisors cite it when sizing the sector, which is why its verifiability became a subject for BIS staff research rather than a vendor dispute.
Examples
Aave is the clearest case of the same protocol carrying incompatible figures. CoinDesk put its total value locked at approximately $25bn across chains when reporting the governance vote of 12 April 2026. The Defiant cited $12.45bn alongside annualised revenue of $402m when Aavenomics 3.0 went into effect on 27 June 2026, and DefiLlama showed $14.254bn on 25 July 2026. Part of that range is genuine price and flow movement over three months. How much is definitional cannot be established from what either source publishes, which is the practical form the BIS paper's verifiability problem takes.
Mantle's own review of the first half of 2026 reported decentralised finance value locked on the network passing $1bn after 230% growth in the half, more than $90m of real-world-asset value locked and over $2bn of community-owned assets. The network's treasury is also its largest allocator: up to $400m of the Mantle Index Four fund, tokenized by Securitize from April 2025, was anchored by the Mantle Treasury. A figure that counts a treasury's deposit of its own assets is measuring something, but not third-party demand.
Risks and limitations
The central dispute is whether the metric should be repaired or replaced. One camp treats double counting as an artefact to be netted out, and proposes redeemable-value measures or verifiable, onchain-only computation. The other holds that layered claims are the product, not an error, and that netting them out understates a system in which the same collateral genuinely does secure several obligations. Both are defensible; what is not defensible is quoting one number as though the question were settled.
Three further limitations bear on use. The measure says nothing about liabilities: a lending protocol with large deposits and large borrowings looks the same as one with deposits alone. It says nothing about revenue, and protocols with similar deposits can earn fees that differ by an order of magnitude. And it says nothing about the terms on which capital would leave, which is the question that matters in stress; deposits attracted by a time-limited incentive are not the same liability as deposits attracted by a rate.
Finally, the price leg depends on an oracle or a reference feed. A measure of size inherits every weakness of the pricing source used to compute it, including the manipulation of thinly traded collateral.
Key metrics
Read total value locked alongside three things that discipline it: borrowings outstanding, so leverage is visible; fees and revenue over the same period, so the deposits are shown to be earning; and net flows in token units rather than dollars, which separates capital movement from price movement. A redeemable measure, where a provider publishes one, is the better headline for sizing exposure. For any single protocol, the useful pair is deposits and the share of deposits represented by receipt tokens issued elsewhere.
Related research
The BIS working paper of 14 May 2025 is the reference on verifiability and the gap between published and reproducible figures. Luo, Feng, Xu and Tasca supply the quantification of double counting and the procyclicality result. For the incentive side, the Uniswap and Aave governance records of late 2025 and 2026 show how fee and buyback decisions are argued in front of deposit figures, and are the best available evidence on how the metric functions inside a protocol's own decisions.