PR #825 follow-up: close consensus shard + snapshot-restore risks

[linear]

Follow-up tracker for risk items left open by PR #825 (snapshot bootstrap + genesis-baked validators + consensus shard enforcement). Parent: DEM-713.

Open items

• Audit production env vars. Confirm every mainnet/staging operator has DEMOS_REQUIRE_VALIDATORS=true set. Currently the default is permissive-with-warning; if any production node misses this, its consensus shard falls back to legacy behavior where any peer can mine. Consider making strict mode the default (or hard-fail at boot when running on a real chain id).
• Close the verify → restore TOCTOU. verifySnapshot and restoreSnapshot run as two sequential async ops over the same filesystem. Adversary model is narrow (operator already trusted, FS write between two ops), but the docs acknowledge the gap. Refactor to verify-then-buffer or verify-then-stream-with-rolling-sha so the bytes consumed by restoreSnapshot are the same bytes hashed by verifySnapshot.
• Byzantine 3-validator rehearsal. Currently deferred because mnemonics for the other two baked validator pubkeys (0x24c664… node3, 0xc8bc58… node2) are not yet available. Was tracked as Mycelium EVM Balances for ETH (and WETH), USDC, USDT #156 (closed/deferred) — need to either obtain mnemonics + rehearse, or scope a multi-mnemonic dev/devnet swap.
• Snapshot distribution path. The ~20 MB JSONL is committed to the repo for reproducibility. Long-term plan was IPFS/CDN distribution post-clone — re-scope when comfortable.

References

• PR: feat(restore): snapshot bootstrap + genesis-baked validators + consensus shard enforcement #825
• Runbook: forking/restore/RUNBOOK.md §2.4 (DEMOS_REQUIRE_VALIDATORS)
• Plan: forking/restore/PLAN.md

Filed by Nesy-Claude per KYN-192.