On-chain data
How protocols and their users behave on-chain: transaction flows, state, and the architecture behind a set of addresses.
Odd Sequence
Web3 research team
We study how on-chain systems behave and where they fail: on-chain data, smart-contract mechanics, AI-driven analysis, and risk in financial protocols. We measure first, then apply it in practice and publish the data and the tools.
The Repeated-Runs Effect
145 runs, 6 tools, 5 contests: one run finds 56% of what a tool can find, two runs 77%, three runs 89%.
pentimento: an open-source AI analysis pipeline for Solidity
MIT licensed. A finding is confirmed only after a proof-of-concept test runs. The verdict is computed in code.
candidate → checks → verdict in code → PoC runs → confirmed
github.com/danogurtsov/pentimento →
Research areas
What we work on
The areas the work covers right now. The list changes over time.
Contract mechanics
How deployed contracts work and where they break: proxies, roles, money flow, upgrade paths, and edge cases.
AI for analysis
What language-model tools can and cannot do for security and analytics, measured on real data.
Risk in financial protocols
Finding and quantifying risk in DeFi systems: oracles, collateral, liquidity, incentives, and failure modes.
Security research
Vulnerability patterns, benchmarks, and how to verify that a finding is real before anyone acts on it.
Selected work
More work
Method
How we work
Measure
We count what exists, run it under controlled conditions and keep the raw results.
Apply
We use the findings on real code and real protocols.
Build
When a tool is missing, we write it in the open.
Publish
We release the result with its method and data, including what did not work.