Thread by @ZeMariaMacedo

At Delphi Labs, we've thought a lot about oracles in the context of our contributions to @mars_protocol and @astroport_fi

In this report, we summarise our key learnings and propose a new framework to set TWAP parameters based on the cost of attack

members.delphidigital.io/reports/attack…

Oracles are key crypto infrastructure, lying at the core of all debt-based protocols such as money markets, derivatives, perps, etc. They're also one of DeFi's biggest attack vectors

This article by @samczsun provides an excellent summary of the issues

samczsun.com/so-you-want-to…

At the highest level, money market oracle manipulation attacks can be broken down into two main categories:

(1) False collateral value increase

(2) False collateral value decrease / debt asset value increase

In (1) the goal is to pump the price of a collateral asset, enabling the attacker to borrow > than the collateral's real value

In (2) the goal is to dump the price of a collateral / pump the price of a debt asset to 'falsely' liquidate another user (i.e. buy collateral cheaply)

While there are various kinds of oracle solutions, for this article we focus primarily on the on-chain TWAP oracle initially pioneered by @Uniswap and used by @astroport_fi

This will also be the primary oracle used by @mars_protocol's Fields of Mars

We analyse the costs of various attacks and use this to derive optimal TWAP parameters

This research has informed design choices on @mars_protocol TWAPs. We hope it can also be useful for others in the space

If you want to talk about this, our DMs are always open

All credit goes to the Delphi Labs risk team, specifically @ayana_eltai and @JonathanErlichL who crushed this research

A special thank you to @euler_mab, @Mudit__Gupta, and many others who provided feedback and helped refine our thinking on this