Writing and published work

Cross-Frontend Attribution Methodologies in Decentralized Exchange Volume

DEX volume is routinely read as a proxy for organic adoption by liquidity-mining programs, L2 foundations and grant committees, and it is a poor one. The paper sets out four independent attribution heuristics that recover a trade's originating frontend from transaction data (calldata suffix trapping, proxy-router detection, multi-hop origin tracing, and fee-recipient clustering), the execution-quality framework that measures realized price against a volume-weighted baseline, and the sampler evidence behind both. It is explicit about the limitation that matters most: the attributable sample is biased, and the paper says which way.

ClearTrace: DEX aggregator execution-quality dataset

Median slippage against a one-minute VWAP baseline, router revert rates, and a slice by originating frontend across Ethereum, Base, Arbitrum and Optimism. The frontend slice is the dimension most public DEX datasets collapse. Free to use commercially with attribution, and mirrored so nobody has to trust a single host.

Three Ways a Number Goes Wrong Before You Publish It

"On a single day in July, three numbers in my company's outbound material turned out to be false."

Three failures that all cleared review, because review catches typos and not derivation: counting database rows as though they were entities, publishing a limit of the instrument as a fact about the market, and reading missing data as a finding. The argument is that a number is not checked once and then true, so the checks belong in a script that runs again rather than in a person reading the draft.

Verify quotes through fork simulation

"In our current sample, about 78,000 fork-verifiable quotes across eleven aggregators between June 23 and August 25 this year, the most reliable venues fail on roughly half a percent of them. The worst sits near 12%, about one quote in nine. Same tokens, same minute, same market."

On why cost and reliability are separate axes that often disagree, why failure rates move with trade size in a way single-size testing hides completely, and the version of the test any desk routing real volume can run on its own executions.

Measure on-chain execution with independent benchmarks

"Because every DeFi trade settles on a public ledger, you can reconstruct the price a swap received and benchmark it against a volume-weighted average at the moment it executed. You can measure quote accuracy and MEV exposure per venue and per frontend, after the fact, without trusting anyone's marketing."

Traditional finance pays for trade surveillance and best-execution reporting and it is still largely opaque to the customer. The argument here is that the same accountability can be a public good on-chain, computed from data anyone can audit, provided price fairness and exposure to predatory ordering are kept as two scores rather than blended into one.

Audit origins, reject vanity metrics

"Honestly, the most effective gamification in DeFi is also the best at fooling the people running it. Points programs, liquidity mining, airdrop farming: they all light up the dashboards. Then the rewards stop, and you find out how much of it was ever real."

A large share of incentive-program engagement is farmers, sybil wallets and bots wash-trading to game the reward. The honest test is retention after the emissions stop, which is the question ClearTrace was built to answer: how much of a program's volume genuinely originated with users, and how much of it was paid for.

Uniswap UniversalRouter V2 and V2.1.1 decoding, four chains

Router deployments that no public model decoded, so the flow through them was landing unattributed in everyone's data, not just ours. The fix belongs upstream in the shared spellbook rather than in a private query, which is where it was submitted. The pull request is open and not yet merged.

ClearTrace research notes

Shorter write-ups that sit between the paper and the dashboard: how the execution-quality numbers are actually computed and why there is no single score, what unattributed volume on Base turned out to be, and how the quote sampler verifies a quote instead of trusting it.