[COMPLETED EXPERIMENT] EIP-8198: schedule-free EL survives falsification

[chugarchugarr]

Result — COMPLETED / NO FALSIFIER

Adversarial falsification experiment for the parked EIP-8198 schedule-driven EL proof.

Question tested:

Can repeated 12s -> 10s -> 8s -> 6s execution invariants be preserved with fork-local EL constants only, without any runtime EL slot-duration schedule?

Result: yes for every invariant exercised here. No falsifier was found.

Exact experiment head: 4cecc230422ad01815c0535d20d877dca4525d34

Focused hosted-runner proof:

• workflow run: https://github.com/chugarchugarr/execution-specs/actions/runs/35881396301
• Ruff check — PASS
• Ruff format check — PASS
• focused falsification suite — 109/109 PASS
• elapsed focused pytest time — 0.45s

Repository-native gates on the same head also passed the available hosted checks, including just static and packaging.

Competitor

fork_local_competitor.py contains one immutable EL parameter snapshot per synthetic fork:

• gas-limit target;
• rational base-fee response coefficient;
• blob maximum/target/update fraction.

It contains no SLOT_DURATION_SCHEDULE, SlotDurationEntry, slot-number lookup, or epoch lookup.

The base-fee constants satisfy the cumulative wall-clock identity without runtime duration history:

k_new = k_old * new_duration / old_duration = k_base * active_duration / base_duration.

For the synthetic eras this produces:

• 12s: 1/8
• 10s: 5/48
• 8s: 1/12
• 6s: 1/16

Gas targets likewise compose as:
72M -> 60M -> 48M -> 36M.

Falsification matrix

The schedule-free competitor passed:

1. exact base-fee equivalence against the schedule oracle across 12s/10s/8s/6s, multiple gas-utilization points, and multiple base-fee magnitudes;
2. composed gas-target equivalence;
3. exact oracle blob constants at every synthetic era using fixed fork-local constants;
4. exact blob-price equivalence at multiple excess-blob-gas values;
5. the discrete blob-rounding probe.

The rounding probe is important: direct-from-baseline fork constants differ from mechanically sequentially rounded constants at later eras, but can be at least as close to the original target/max throughput objective. Integer blob discreteness therefore does not, by itself, force the EL to carry runtime duration history.

Consequence

This experiment falsified the stronger claim that the current evidence requires a runtime EL slot-duration schedule.

The parked schedule-driven implementation remains useful as:

• an executable oracle;
• a repeated-era composability proof;
• a regression/falsification harness.

It is not presently established as necessary execution architecture.

Reactivation condition remains narrow: an upstream EL implementation/test must expose an invariant that valid fork-local constants cannot preserve, or an execution rule must genuinely require historical duration-era knowledge at runtime.

This PR targets only this fork. No Ethereum upstream repository was modified.