Recently, I noticed a pretty interesting DeFi phenomenon. Previously, a Solana developer used 11 pseudonyms to stack protocols on-chain, artificially inflating Solana’s TVL; later, it was exposed when DeFi Llama changed its calculation method. This incident made me start to wonder: how much “water” is there in the TVL metric?

To be honest, TVL (Total Value Locked) looks pretty straightforward. The more money a project locks, the bigger its scale. Dividing market cap by TVL can also be used to estimate valuation. But the problem is that this number can be played with. It’s a static snapshot—TVL fluctuates a lot. Short-term incentives from the project team and token price volatility can both cause TVL to change dramatically. More importantly, the TVL of different projects simply can’t be directly compared.

In foundational protocols like DEXs and lending, TVL is still relatively reliable. Uniswap’s TVL is just liquidity—no tricks. But for things like Curve and Sushi, which have governance tokens, users stake tokens to receive fee-sharing, and that part is listed separately as Staking, so the calculation is different. Lending protocols are even more complex: Compound’s TVL is the gap between total deposits and total loans (total deposits minus total loans). Aave allows staking AAVE tokens and LP tokens. MakerDAO, meanwhile, is simply total deposits.

The real trouble comes from projects built on top of other protocols. Yield aggregators like Yearn and Convex Finance put users’ funds entirely into underlying protocols for yield farming, which leads to double counting. Convex locks a large amount of CRV on Curve to help users earn higher returns. When Solana’s TVL was only $10.5 billion, Saber and Sunny together accounted for $7.5 billion—how exaggerated that proportion is should be obvious.

Liquidity staking protocols are also a big pitfall. The stETH derivatives issued by Lido have 21.6% used as collateral on Aave, and 14.7% provide liquidity in Curve’s ETH/stETH pool—these are all double counted. Middleware tools like Instadapp are even more extreme: they’re basically a front-end layer that helps users manage various DeFi positions, but the funds are already deployed in other protocols. At its peak, TVL was $13.5 billion; now it’s $2.6 billion—there’s no reason it should be counted as base-layer (public chain) TVL at all.

The key turning point was when DeFi Llama changed its methodology and, by default, removed the double counting caused by protocol stacking. After this change, on-chain TVL data dropped significantly, but it also became more accurate. Now, even though the DeFi TVL numbers are smaller, the “value” is higher.

So the conclusion is simple: don’t be fooled by TVL. At the application layer, it can be used for cross-project comparison, but you must understand what each project’s TVL actually represents. At the public-chain layer, after the bubble bursts, the data becomes more valuable instead. Recently, I’ve also been tracking some related assets on Gate—if you’re interested, you can check them out yourself.