EIP-8372: Normalized state gas limit

⌬LLM No opcode is introduced.

⚇Human No rationale recorded.

⌬LLM No opcode result or non-gas behavior is changed. Any effect of the new CPSB on state-creating opcodes is a gas change, which this anchor explicitly excludes.

⚇Human No rationale recorded.

⌬LLM No precompile is introduced.

⚇Human No rationale recorded.

⌬LLM No existing precompile logic or gas schedule is modified.

⚇Human No rationale recorded.

⌬LLM No system contract is introduced.

⚇Human No rationale recorded.

⌬LLM No opcode's state access, gas-charge position relative to access, or definition of a recordable access is changed.

⚇Human No rationale recorded.

⌬LLM No blob-gas counter, price, target, limit, or accounting rule is modified.

⚇Human No rationale recorded.

⌬LLM The block-level state-gas budget is modified and a normalization is added. This is the rubric's score-2 case. The transaction reservoir and its spill path into execution gas are expressly retained, so score 3 is not warranted.

⚇Human The block-level state-gas budget is modified (scaled to a share of the block gas limit) and `CPSB` recalibrates with it; the spill path and charging sites are untouched.

⌬LLM The proposal adds no refund or refill mechanism and does not modify the inherited refund rules.

⚇Human No rationale recorded.

⌬LLM No transaction type is introduced.

⚇Human No rationale recorded.

⌬LLM Existing transaction types receive a changed inclusion-validity check that affects existing state-gas cases. Updates are limited to expected arithmetic and boundary cases and require no transaction format or testing-infrastructure redesign, matching score 2.

⚇Human With a scale below 100, the EIP-8037 admission rule (`tx.gas <= state_gas_available`) silently caps every transaction's gas limit at the scaled share of the block gas limit — an unstated consequence, and the dominant blast-radius mechanism until tooling learns the bound (31,873 fixture executions in the prototype's first fill). Existing validity tests need limited, mechanical updates.

⌬LLM No block or header field is added.

⚇Human No rationale recorded.

⌬LLM No transaction, block, header, receipt, or interface encoding is changed.

⚇Human No rationale recorded.

⌬LLM No new block RLP validation or encoding mechanism is introduced, so the rubric's syncing-specific anchor is not triggered.

⚇Human No rationale recorded.

⌬LLM Activation selects protocol constants and begins applying new rules; it does not modify state or an existing internal variable at the activation block.

⚇Human No rationale recorded.

⌬LLM No Engine API endpoint, field, or communication mechanism is added or modified.

⚇Human No rationale recorded.

⌬LLM The transition behavior changes, but the proposal specifies no new field or mechanism that must cross the transition-tool interface.

⚇Human No rationale recorded.

⌬LLM Existing affected tests must change expected values and validity outcomes, rather than mechanically add an assertion for a newly produced artifact to otherwise unchanged tests.

⚇Human No rationale recorded.

⌬LLM The mechanism touches a limited set of critical accounting and validation components and changes their invariants, requiring targeted review, boundary testing, and fuzzing. It does not introduce the broad multi-component security redesign required for score 3.

⚇Human The arithmetic lands in a consensus observable: header `gas_used` is recomputed through the normalization fold, so a rounding or crossover divergence between clients is a chain split, warranting targeted review and fuzzing across the boundary ranges. Parameter miscalibration remains the economic failure mode, fixable only by a later fork — flagged by the EIP itself.

⌬LLM Tests need multiple boundary families: floor-division remainders, exact-limit and one-over-limit raw state gas, exact normalized block limit, and the execution/state max crossover. Their interaction requires an elevated set of cases rather than isolated single boundaries.

⚇Human The scaled-capacity admission boundary, integer normalization rounding, and the dominant-dimension switchover are each boundary-prone, and the switchover needs an elevated case count: the `max(execution, normalized)` crossover must be pinned at equality and one raw unit either side, non-divisor scales collapse rounding ranges of raw values into one normalized value, mid-transaction reservoir-to-spill crossings move with the grant, and each boundary is observed twice (raw receipts, normalized header).

⌬LLM No cryptographic primitive, algorithm, proof, signature, hash, or associated validation behavior is introduced or modified.

⚇Human No rationale recorded.

⌬LLM EIP-8372 depends on and modifies EIP-8037, so coordinated state-gas and block accounting tests are required. The interaction is deep but limited to one directly required EIP. EIPs 7999 and 8075 are contextual alternatives or analogues in the sealed text and are not counted as co-implemented interactions.

⚇Human A delta on EIP-8037's block accounting, interacting with EIP-7825 (the admission-cap consequence, measured) and EIP-7778's block-accounting suites — limited, mechanical coordination.

⌬LLM Clients cannot baseline concrete consensus vectors until the two localized parameters are agreed. The formulas otherwise determine behavior, so this is localized agreement rather than newly observable unspecified semantics requiring repeated broad re-baselining.

⚇Human Two consensus-relevant behaviors are unstated in the EIP but localized: the sub-100-scale admission cap (needs the comparison re-anchored or the ceiling stated) and receipt gas totals staying raw while the header normalizes. Clients need agreement on both before vectors are baselined; the TBD constants are churn already counted under patterns.