The Herd team will be in New York during ETHConf June 8-14, we'll be co-hosting an event at Chelsea Piers on the 9th. RSVP here!

The Death of One-time Token Due Diligence

Tokens are increasingly becoming the centerpoint of complexity instead of protocols, which have become more modular and configurable. The surface area of a given vault can easily balloon to 300+ contracts with constantly changing operations across underlying tokens (rotating EOAs, upgraded contracts, new bridges, etc).

In the first five months of 2026, we've seen the consequences of this bloated operational complexity with almost a billion dollars worth of hacks. Notable hacks include Resolv in early March (USR mint hack), then Drift (multisig hacked) and KelpDAO (rsETH bridge hack) in April. Hackers would then route the hacked tokens through other protocols to gain exit liquidity, creating contagion affecting billions of dollars of other tokens in places like Aave.

Throughout this difficult period, two questions have been asked by everyone:

• How can I continuously due diligence these tokens, instead of one time reviews?

• What is the contagion to my portfolio from a market shock/hack on another token?

This can be measured with the right risk models, which are typically required and regulated in traditional finance for examinations and reporting. However, we don't have any standards or legal frameworks for this in defi yet. Most the tooling we have has traditionally been security focused, with transaction/address screening products. Some expert teams have worked with protocols like Aave on one-time research and analysis of tokens, but this is no longer enough.

You can think of modeling risk like navigating a map, where factors like construction, weather, and traffic conditions all affect your decision making. You could go off of just what you see and feel on the road, but your visibility into the pathways will be poor. We decided to start out by building out the token map. But every token is structured completely differently - there is no "ERC20" for all the composable dependencies of a token. So, we leveraged our own MCP/CLI and built a research harness around it that can map out and maintain an "asset liability graph" for any token.

Herd's Token Balance Sheet Graph

Every balance sheet has two sides, assets and liabilities. We define assets as all the positions backing a given token. Liabilities are the issued token itself, and what protocols it's being used in and re-wrapped.

This approach isn't inherently novel, and a version for MakerDAO/DAI (now Sky and USDS) has been written about before. Having a balance sheet modeled out is required to then model common metrics such as a Liquidity Coverage Ratio (LCR) and Capital at Risk Ratio (CARR). The Federal Reserve then comes in to test bank models with Comprehensive Capital Analysis and Review (CCAR), setting different stress assumptions each year.

Our key difference in approach is that we recursively breakdown every position (line item) to get its underlying positions and who manages them to create a balance sheet graph. Now, let's look at an example graph we built with our agent.

iUSD is a yield-bearing stablecoin issued by InfiniFi. Here's a portion of their token graph:

Each graph also comes with an aggregated balance sheet view that can be exported as a json/csv/pdf. All data can be queried live, and the agent has many monitors set for continuously researching changes in underlying positions/managers.

Both sides of the balance sheet are sources of contagion risk. On the asset side, ~43% of iUSD's reserves are off-chain Fasanara Capital positions. If there's a credit event, those can't be liquidated onchain. On the liability side, ~52% of the supply is re-deployed as siUSD into Morpho, Pendle, Silo, and Euler, with another ~33% locked in weekly tranches (liUSD-1w through 13w). A freeze on the assets doesn't stay contained, it creates redemption pressure that cascades through every protocol holding siUSD downstream. Our graph shows you exactly where the shock enters and how it travels.

Building Transparent Risk Models For Every Token

To create risk models, you need to simulate onchain functionalities across contracts with different actors. The balance graph gives us the starting "map" to explore, but it doesn't capture all possible paths and counterparties involved. Our agent can double click into each node and edge in the balance sheet graph and further research sets of functions for managing everything from deposits, redemptions, and configurations.

Here's a function graph that shows who can call the mint function on USDC on ethereum (a few EOAs and then a few Circle bridge contracts):

Together, the balance sheet and function graphs give you everything needed for an accurate exposure risk model. Both need to stay live as positions get rebalanced, contracts are upgraded, new managers are added. The agent monitors and maintains every layer continuously, flagging changes and tweaking model assumptions as they happen.

Our product directly answers both earlier questions. For continuous due diligence, the graph is maintained automatically by agents with human review - no more one-time reviews that are stale the next week. For contagion, when KelpDAO's rsETH bridge was hacked in April, any token with rsETH in its balance sheet graph had a traceable exposure. With the graph you see immediately whether that shows up as a liquidity crunch during a run, bad debt for which depositors, or cascading liquidations as sell pressure pushes correlated collateral below LLTV.

Work with Herd

At Herd, we've worked closely with vault curators like Steakhouse and Chaos Labs using our Explorer and MCP/CLI. This new product is the next evolution of leveraging AI to manage the complexity of crypto at scale while remaining accurate.

We're currently working with a few design partners on custom risk models for their tokens and portfolios. We'll have more to reveal soon and are giving out beta invites to anyone actively issuing or managed tokenized assets. Subscribe to this newsletter to get our updates, I'll be posting research into interesting tokens each week as well.

If you want to model token/portfolio risks or have ideas for working together, please reach out to andrew@herd.eco (andrewhong5927 on twitter/tg) for a product demo.