data sources & limitations

mev-inspect-py (by Flashbots) is used by zeromev for MEV classification.
as such much of the information from the MEV-Explore data & metrics page applies here.
due to known issues with sandwich profit estimation in mev-inspect-py, we recalculate sandwiches using our own AMM pool extraction method.
other MEV amounts are sourced directly from mev-inspect-py but with dollar exchange rates for these instances recalculated by zeromev using on-chain Uniswap v2/v3 swap data.
because of this, our USD denominated results may differ from those found in MEV-Explore which uses external data sources.
known issues with mev-inpsect-py can be tracked on the Flashbots github.
notably, there is a problem with detecting split arbitrages which may contribute to the disparity in arbitrage results below.
there are also issues with overlapping MEV (eg: arbs co-mingled with sandwiches, and sandwiches that crossover with each other) which no MEV detection software fully handles at present.
as such, this data must be considered a lower bound of MEV.

data source comparison test

Eigenphi provide a good data source comparison test set as they employ a very different MEV classification methodology to mev-inspect-py and zeromev
as can be seen below, the sandwich attacks recalculated by zeromev produce similar results to Eigenphi.
mev-inspect-py arbitrage classification is not yet recalculated by zeromev and misses around 64% of instances compared to Eigenphi at time of writing.

data will be subject to change as MEV classification improves.
please bear this in mind when referencing our data and consult our change log for details of any revisions.