Paragon Labs

Research

Five areas of research.

Some of it is contracted. Much of it is self-directed. Each area ends in something that can be checked: a dataset, a proof, a paper, or working software.

01

Market microstructure

We measure how markets behave at the level of individual orders: who pays fees and how much, what stale quotes cost liquidity providers, and how execution quality changes with market conditions.

  • Fee spend across perpetual futures markets
  • Stale-price and latency arbitrage
  • Order flow and execution-quality datasets

02

Hyperliquid

Hyperliquid runs an order book (HyperCore) and an EVM (HyperEVM) on one chain. CoreWriter lets contracts on one act on the other. We study how the two interact, and publish data research on its markets.

  • CoreWriter: asynchronous execution between HyperEVM and HyperCore
  • Programmable self-custody: account and permission models for automated execution
  • Fee-spend and microstructure data on perpetual markets

03

Intents and MEV

In intent-based systems, users state an outcome and solvers compete to deliver it. We research how these systems price, route and settle, and how value moves through transaction ordering.

  • Solver competition and quote design
  • Signed-quote and on-chain liquidity in one venue
  • Work alongside CoW Swap and independent solver teams

04

Mechanism design

Different assets need different market rules. We design mechanisms from the asset outward: its redemption process, its settlement delay, and the information available when it trades.

  • Modular exchange architecture: pricing, fees and access as separate parts
  • A marketplace for liquid staking, where the redemption queue is a price for time
  • Dynamic fees driven by inventory and quote age

05

Frontier tooling

We’ve been early partners to research organisations building frontier technology stacks.

  • Collaboration with Nethermind on a system that compiles EVM software into Lean, making formal verification of blockchain code a developer-ready stack. This work makes code safer.
  • Early development partner of Starknet, building on the Cairo v0 language and helping mature ZK-provable language design in its early days. This work makes security-sensitive code run privately.

Prior work

Machine learning

Before Paragon Labs was founded, a member of the team published research on fast uncertainty estimation for computer vision (CVPR 2020).

Original paper on arXiv

Papers

6 papers