CHAINWRIGHT / NATIVE BENCHMARKS / 2026-10-29

21 fresh synthetic native-chain runs, three repetitions for each case. Previous shipped EVM examples are the baseline. The only optimized source change is a constructor-fixed authority made immutable. Return/revert/state assertions and ABI comparisons pass for the listed inputs; this is not a security audit or universal equivalence proof.

Read summary.json, receipts.json and sources/. run-server.py expects the already installed isolated Linux runtime at /opt/chainwright-studio with the pinned Agave2.3.13/platform-tools1.48/Anvil1.8.6 and existing bwrap/runuser/prlimit. It makes an isolated copied runtime; it does not update the public broker. Do not install or activate a public service by running this archive. Preserve its authentication, quotas, deadlines and network sandbox. Node dependencies and native toolchains are separate source/runtime prerequisites in the Chainwright source handoff. No credentials or real-wallet material is included.

Solana adapter output is compared with the same pinned TypeSOL0.24.0 build function. C, IDL and SBF hashes match for all three examples. Synthetic actors are seeded only in the benchmark copy to keep PDA bumps stable. The public runner still generates fresh synthetic actors.

The two EVM source/ABI comparisons preserve tested observable outputs, not raw storage layout or creation/runtime bytes. Bytecode size grows. Hook CREATE2 addresses and calldata bytes differ. Hook callback receipts did not improve; the deployment receipt did. The separate diagnostic records report Anvil's Cancun node setting, no external fork, and observed gas. Do not infer mainnet costs, EIP activation or opcode-level savings from those diagnostic fields.

No real wallets, paid provider, external RPC, live protocol deployment or token issuance. Original source is MIT licensed. Compiler/runtime/SDK upstream licenses remain in the separate exact source package and distribution notices.
