Add EIP: Fixed-Cutoff State Tiering

[weiihann]

Need to merge #11784 first

[eth-bot]

File EIPS/eip-8296.md

Requires 1 more review from Editors: @g11tech, @jochem-brouwer, @lightclient, @samwilsn

[github-actions]

The commit 717a336 (as a parent of 473103b) contains errors.
Please inspect the Run Summary for details.

[CPerezz]

Per EIP-8188, several operations bump more than one leaf, but EIP-8296's cost table has one row per leaf type and says the operation is charged its normal gas plus the corresponding inactive surcharge (singular).

Three concrete cases are undefined:

1. SSTORE to a dormant contract. 8188: an SSTORE bumps the slot's last_written_block and the account's (storageRoot recomputed). If both slot and account are Inactive, does the caller pay INACTIVE_STORAGE_WRITE_SURCHARGE only, or storage + INACTIVE_ACCOUNT_WRITE_SURCHARGE?
2. Value transfer between two dormant accounts. 8188: a balance transfer sets last_written_block for both sender and receiver. If both are Inactive, is that one account surcharge or two?
3. Direct collision with State Creation Is Not Tiered.
   Creating a new slot (SSTORE 0→x) in a dormant contract is exempt from the surcharge (it's creation, priced by 8037) — but it cascades an account storageRoot mutation, and that account already exists and is Inactive. By the literal rule "tier is evaluated for mutations of state that already exists in the trie," the cascaded account bump would be surcharged.

So "creation is not tiered" and "Inactive account mutation is surcharged" contradict each other for this very common case.

Extra stuff worth signaling:

4. Another topic is how intrinsic gas should be affected based on this PR.
5. EIP-8037 (same family, explicitly referenced here) splits gas into regular-gas and state-gas, with different block-accounting and a different relationship to EIP-7825's TX_MAX_GAS_LIMIT. EIP-8296 says the surcharge is "consumed like ordinary execution gas," "not refundable," "consumed even on revert" — which reads as regular-gas / single-dimension. But it never reconciles with 8037's two dimensions. Since reviving cold state is I/O work (not state growth), regular-gas is the likely intent — but with 8037 active, this must be explicit, or block-level accounting is undefined.
6. From 8188, legacy (pre-fork, never-rewritten) accounts/slots decode with last_written_block = 0, which is < CUTOFF_BLOCK for any valid cutoff ⇒ Inactive. That's exactly the intent, but 8296 never says it. Also, the inverse edge: state created in [8188_activation, CUTOFF) is Inactive, so its second write pays the surcharge even though it was created recently. Both are consequences worth one explicit sentence each so implementers and users aren't surprised.