Same trade, same instant, every configured aggregator — then the returned calldata is
executed in simulation through two independent engines and ranked by what it actually
delivers, not what it advertises. Wrapped native → the chain's dominant stable, canonical
~$1–2k notional. Reading rules in the notes at the bottom; collection code:
workers/latency-bench/swap-bench.
Loading…
Each provider's cross-chain median gap (HyperEVM excluded — thin sims), split into its published fee at standard tier (from the route99 cost model: 0x 15 bps, LiFi 25 bps, 1inch 30 bps; others publish 0 bps per-swap) and the routing residual. A residual near zero means the gap is the fee — the routing itself is competitive.
Loading…
A nonzero point means the same request, at the same instant, got a materially different quote depending on the datacenter asking. Spikes are events; hover for values.
Loading…
Per chain and provider: simulated ok · reverted · errored · quoted-only. Hover a bar for the dominant revert reason — on HyperEVM that is the simulation environment's missing oracle precompiles, a simulator fact.
Loading…
Median quote latency from the standing edge series (7-day window, log scale) against the delivered gap to best from this run. Down-right is slow and worse; up-left is the place to be. One dot per provider × chain.
Loading…
Wei-exact agreement between Nordstern's simulator and Enso Quoter per provider × chain. This audits the instruments, never the providers: Velora's red row is a documented gap in the second engine; its rows stand on the primary verdict.
Loading…
Per-unit delivered output at ×1 / ×10 / ×100 the canonical input, as impact vs each provider's own ×1 (bps). Flatter is deeper routing.
Loading…
Loading…
Loading…