[sim] Run the environment on a real L2 (cheatcode-free mode)

[adachi-440]

Motivation

A code audit shows the environment is structurally portable to a real L2 (public testnet or dedicated rollup, running our own deployed venue clones): the deployer, env contracts (PriceFeed / mock oracles / vuln factory), flow bot + relay, GMX keeper (realtime path), default-profile oracle updates, and scoring reconstruction all use standard signed txs / archive reads. What blocks a real-chain run is a small, well-localized set of anvil-only mechanisms.

Work items

(1) Funding without cheatcodes — fundWallet uses anvil_setBalance (ETH) and dealErc20 (anvil_setStorageAt, stables/bases); admin/keeper funding and vuln-pool window funding are the same. Replace with real transfers from a treasury EOA — tokens are our own MockERC20s (deployer holds mint rights), so mint+transfer is mechanical.

(2) Block-production and ordering assumptions — drop setIntervalMining / evm_setAutomine (the sequencer produces blocks), but the real design question is the loss of anvil --order fees: the default profile relies on fee-priority ordering to land oracle updates at txIndex 0. On Arbitrum-style FIFO/timeboost sequencing that guarantee disappears. The economicGas profile was designed to remove this assumption but finalizes prices via storage-writes, so it is unusable on a real chain — this needs a tx-based redesign (e.g. commit oracle price for block N in a tx during N−1).

(3) Run lifecycle without reset — resetFork (anvil_reset / evm_snapshot+evm_revert) has no real-chain equivalent; a run becomes fresh-deploy-per-run (ERIS_SKIP_RESET semantics).

(4) Mechanical ports — GMX vendor deploy hardcodes hardhat --network localhost + chainId 31337 (edit vendor network config, re-point deployments/<net>); drop the Aave fork-timestamp increaseTime warp (fork-skew workaround, irrelevant on a native chain); scoring requires an archive RPC for historical multicalls.

Notes

• Target is our own closed-economy venue deployment, not real protocols — the fair-price/oracle-injection design assumes the env owns every contract.
• economicGas=false (default) already keeps the entire per-block oracle path tx-based; keep it that way for real-chain runs.

[adachi-440]

Target-chain update: an OP Stack chain we operate ourselves is now the leading candidate (#35 tracks chain construction / approach selection). This materially changes the weight of the work items above:

• (1) Funding — the treasury bootstrap problem disappears: genesis allocs prefund a treasury EOA (or the admin/keeper/flow wallets directly). The remaining work is only swapping fundWallet's cheatcode calls for real transfers/mints.
• (2) Ordering — this item was written assuming Arbitrum-style FIFO/timeboost, where fee-priority ordering disappears. The OP Stack sequencer (op-geth) builds blocks in effective-tip order by default, so the oracle-at-txIndex-0 design of the default profile survives largely as-is. The item shrinks from "tx-based redesign" to "verify op-geth ordering under our tx patterns" (tracked in [infra] OP Stack chain for real-L2 Eris runs: approach selection & Eris-side interface #35).
• (3) Reset — on a self-operated devnet, per-run chain restart from genesis is cheap and preserves the clean-slate-per-run model; fresh-deploy-per-run is only needed on a persistent chain.
• (4) Mechanical ports — chainId / network entries (incl. the GMX vendor hardhat config) become concrete once [infra] OP Stack chain for real-L2 Eris runs: approach selection & Eris-side interface #35 fixes the chain parameters.

Interface between the two issues: genesis allocs, sequencer ordering config, chainId/block time come from #35; this issue consumes them via a new external-chain run mode + deployer network entry.