[kyscott18]

Summary

• Spec is a Monad hardfork or an Ethereum one (Merge..Osaka). When it isn't pinned, the chain ID picks it: 143/10143 use monad-revm's schedule; 1, Sepolia, and Hoodi use Ethereum's activation timestamps (checked against alloy-hardforks). The harness picks monad-revm or mainnet revm through a small Backend trait, and the network stays fixed after init.
• Simulated gas_limit is now max(gas spent before refund + reserve, EIP-7623 floor). Monad has no refunds, so tx_gas_used was enough there. On Ethereum it under-provisions any tx that clears a slot (new reserve_covers_refunds test).
• The sim draws chain.network. Anvil keeps its chain ID (31337), which has no hardfork schedule, so the sim pins the hardfork to match: --network monad runs MonadTen, and --hardfork osaka --enable-tx-gas-limit with 12s blocks runs Osaka. Choosing the hardfork from the chain ID is covered by the Rust and TS unit tests.

Validation

• Rust tests: 9/9. typewriter-evm TS tests: 34/34. typewriter unit tests: 97/97. Typecheck, clippy, and biome are clean.
• Reorg-free seeds, run one at a time:
  • Ethereum: 10/12 pass. 0x4d and 0x4 fail with intrinsic gas too high -- tx.gas_limit > resolved tx gas limit cap.
  • Monad: 6/6 pass (0x11 0x41 0x23 0x74 0xcd 0x16f).

Ethereum blockers this exposes (not fixed here)

1. Osaka per-tx gas cap (2^24 ≈ 16.8M). A backlog settles in one tx; seed 0x87 simulated 486 mutations at 32.7M gas. This is the existing TODO(kyle) cap a submission. Monad has no such cap.
2. EXPIRATION_MARGIN = 5s. It is shorter than one 12s slot, so 1–15% of mutations per seed are dropped from reverted settlements. With expirations off, 0x4d and 0x87 drop 0.
3. eth_sendRawTransactionSync vs the 5s REQUEST_TIMEOUT. With 12s blocks, every sync send times out before inclusion. The sim's transport doesn't resend a known tx, so it passes; on a real node this depends on how the node answers a resent known tx.

🤖 Generated with Claude Code