GitHub
Live probe study

EVM-compatible isn't EVM-equivalent.

We tracked read-only probe cases across EVM chains from . Valid precompile vectors, block clocks, randomness, gas costs and lifecycle boundaries returned different results; every matrix cell links to its raw response or recorded endpoint limit.

Chains
Probe cases
Confirmed differences
Latest run
Every probe on every chain
Mechanismeth_call, no to

How a probe works

A probe is a few bytes of contract init code. Sent to eth_call without a to address, the node runs it exactly as it would run a deployment, then hands back whatever the code returns. No transaction, no signature, no deploy.

    Headline results

    Five results that change a multi-chain review

    Each card shows what a developer would reasonably assume, what the probe returned, and the change that makes the code hold up on every chain.

    Loading findings…

    Second sweep

    Second sweep: gas, real inputs and limits

    1. Loading…
    Centerpiece

    Chain × probe matrix

    Each cell compares one chain's answer with Ethereum's for the same probe. Pick a cell to see both raw values and the file they came from.

    Scroll sideways on small screens, or open one chain
    Gas schedule

    Same opcode, different price

    Hover or tap a square for both values and the raw files

    Per chaindiff against ethereum

    Pick a chain, see what differs

    Everything on one chain that answered differently from Ethereum, side by side with Ethereum's answer.

    CatalogCATALOG.md

    Every catalogued finding

    Measured differences, plus source-verified behaviors that need a real transaction to observe. Each row links its evidence.

    Before you deploydocs/checklist.md

    What to check before you deploy

    Start from the target chain. Attach the RPC, block and raw response to each answer. Ticks stay in this browser only.

    Methodread-only rpc

    Method and limits

    Independent re-checks of random catalog samples

    RPC endpoints used

      What read-only probes can't measure

        This research was run with AI agents and may contain mistakes. Every claim links its raw evidence, and corrections are welcome.

        Prior art

        Run it yourself

        The probe kit runs every check with one command and can sit in CI, so a release fails when a known answer changes on your target chain.

        ./probes/run.sh