Retire fork code. Keep verifiable history.

Loading evidence

Each upgrade adds another set of rules to maintain. Base Era moves historical execution into pinned workers, creating a path to remove superseded logic from the active client—not stop verifying old blocks.

Technical report ↗ Raw evidence ↗ Deletion experiment ↗

A real deletion diff: ~6k Rust LoC

Full diff + per-file counts ↗

5,907net Rust lines removedlatest-only source experiment

2,743production Rust linesnet reduction

3,164test + benchmark linesnet reduction · counted separately

Fresh Base clone, Beryl-and-later horizon: 6,378 Rust lines deleted, 471 added. Physical lines include comments and blanks; bytecode and build metadata are excluded. This is a separate source experiment, not a change to the live demo.

Retires old channel/batch pipelines, upgrade transaction builders, fee and receipt rules, and historical execution/proof assembly. Current rules, old-data decoders, schedules and future forks stay. 648 tests pass; 8 documentation examples remain ignored.

Keep the latest client small as the chain accumulates forks. Pin an era → route and verify its history → migrate remaining consumers → delete superseded host bodies. Repeat at each fork. Realizing the full estimate requires historical derivation and proof dispatch, beyond this demo’s execution worker. That integration adds code; the deletion candidate is not safe for historical replay.

A historical call, through the worker

Captured request + logs ↗

Loading recorded local devnet call…

HTTP round trip

—

Worker exchange

—

One recorded warm sample, not a live dashboard. Exchange includes IPC, state reads and execution; remaining round-trip time is unattributed. Capture method ↗

Loading acceptance results…

Performance

History host + worker vs original reference

Measured performance on the same local devnet workloads
Metric	History host	Reference
Loading measurements…

Warm medians, 20 samples. Hardware, build profiles and CPU/memory measurements are in the linked report.

223× faster historical calls; 254× faster estimation. Versus the original debug-build demo, including optimized compilation. Persistent isolated workers cache configuration, never execution state; warm request frames shrink from 9.37 MB to 2 KB. Calls remain 51× slower than the optimized in-process reference. Measurements + limits ↗

Execution boundary

Isthmus activation: block 20

Execution routing across the activation
Blocks	Rules	Executor
1–19	Holocene	Pinned worker
20	Isthmus cutover	Host; old-era parent
21+	Isthmus	Host

Real fork differences: upgrade deposits, GasPriceOracle flag, and requests/withdrawals commitments.

Protocol v2: reusable sessions; every request binds chain/config, parent, candidate, operation and artifact digest.
State: immutable parent reads in; account/storage/code deltas and era-correct receipts out. Only the host writes the DB.
Isolation: SHA-256 approval, sealed executable, Landlock + seccomp. Failures never fall back to latest rules.
Independent builds: separate lockfiles; worker Alloy 2.4.2 vs host 2.4.1. Reth delta: 13 files, +300/−42 lines.

Coverage: sequencer/verifier parity, restart, cross-cutover reorgs, malformed Engine/import blocks, historical calls/estimates/traces and worker failure recovery. Additional tests: 13 host, 18 worker, 9 artifact-integrity.

Reproduce

just demo

From a clone of this repository: builds both implementations, runs the devnet and acceptance suite, then exports evidence. Leaves the network running and prints endpoints and a stop command. Prerequisites + build guide ↗. Linux source-build demo; no binary release yet.

Ethereum context

EIP-4444: History retention. Its full-sync design already proposes stitching historical execution engines together; Base Era explores that approach.
State expiry: Inactive state can expire and be resurrected. Distinct from history pruning or witness-based statelessness; not implemented here.
Era / Era1: Archive data formats, not executable fork rules. This demo does not read or write them.