New gas proposal

Worst-case-across-clients gas costs derived from the runtime fits, compared against the current Osaka cost table. Rendered from the generated proposal report; see the runtime and glue pages for the underlying per-fit detail.

Generated 2026-07-21 12:25:26Z · fork osaka · anchor_rate 100 Mgas/s

Summary: 12 parameters proposed — 10 increased, 2 decreased, 0 new, 0 unresolved · 4 warnings · 0 poor-fit selections

Proposed gas parameters

Client comparison

Worst client vs. second-worst client per gas parameter. The Ratio column is worst gas / second-worst gas — values close to 1× mean the worst case sits next to the rest of the field, while large ratios flag the worst client as an outlier.

Per-client proposed gas for each parameter. Cells are colored by log2(proposed / current) — red means the proposal is more expensive than the current gas cost, green means cheaper, and white sits at unchanged. Annotations show the absolute proposed gas value; blank rows are parameters with no prior baseline (see warnings below).

Worst-case provenance per gas param

One collapsible block per gas parameter showing every per-client candidate that the worst-case selector saw. Rows are model combos (the source regression's test_name, target_opcode, model_coef_name, and any model_by factors — components constant within a parameter are dropped from the label). Cells carry each candidate's proposed gas; the cell the per-client selector picked is outlined in black. Colors are log2(proposed / current) against that parameter's baseline on a per-parameter symmetric scale.

Warnings

Missing parameters

None.

Incomplete client coverage

Clients with no estimations at all: reth. These configured clients produced no fits for any gas parameter — check that the runtimes CSV contains their rows and that the fixture-name conventions match. Inspect the evm_gasfit warnings in meta.json for the cause.

Missing glue adjustments

Other

• derived: 'REFUND_STORAGE_CLEAR' shadows a raw fork field on 'osaka'
• derived: 'TX_ACCESS_LIST_STORAGE_KEY' shadows a raw fork field on 'osaka'
• derived: 'TX_ACCESS_LIST_ADDRESS' shadows a raw fork field on 'osaka'

Poor-fit selections

Rows where the winning fit's p-value exceeded modeling.poor_fit_p_value_threshold (0.05) or its R² fell below modeling.poor_fit_rsquared_threshold (0.5). The failing threshold(s) are noted alongside each row; selections in ### Winners with poor fit still drive the proposal, while ### Other weak candidates lists losing candidates that the selector dropped in favor of a qualified alternative. See runtime_estimation_autogenerated_report.md for per-fit runtime_ms, pvalue, and rsquared metrics.

Winners with poor fit

None.

Other weak candidates