Backtest

Corrected. Each trade is re-quoted at the block before it executed, because its own swap moved the pool it landed in. Transactions containing more than one Swap log are discarded entirely — a single leg of somebody else’s multi-hop route is not a complete trade, and comparing our whole route against one leg of theirs would flatter this project enormously.

Not corrected. The comparison is gross of gas on both sides. We do not know what they paid, and our own extra-hop cost is not netted out either, which if anything favours them. We also cannot see why they routed as they did: a trade that looks beatable may have been a deliberate venue choice or an MEV-protected order.

The sample is small and recent. Public RPC serves a few thousand blocks, so each run samples the last few hours. Depth accumulates across scheduled runs. Every figure here is recomputed from data/backtest.jsonl, which is committed — so any claim is auditable in the diff that introduced it.