# Svalbard Interop — Closing Session Summary

**Date:** 2026 / 05 / 01
**Hosts:** Tim Beiko
**Time:** 16:00

## 1. Objectives of the Week

The interop week was structured around reaching alignment on three core Glamsterdam deliverables:

1. **EIP-7904** — General opcode/precompile repricing
2. **EIP-8038** — State-access repricing (storage, account access, account writes)
3. **EIP-7732 (ePBS)** — Enshrined Proposer-Builder Separation devnet validation
4. Establishing a **conservative public floor for the post-Glamsterdam gas limit**

Stretch goal: **EIP-7805 (FOCIL)** initial multi-client implementation.

## 2. Devnet Status (Barnabas)

### bal-devnet-6
- Launched at ~06:00 the morning of the closing session.
- Most EL clients onboarded; remaining teams can request inclusion.
- Continuous fuzzing throughout the day; only minor missed slots observed.
- No remaining major fork disagreements — network considered **stable**.

### ePBS devnet
- Built on top of the same EL branches as bal-devnet-6 (BAL devnet-6 EL branches + runs-tarnet-2 CL branches).
- Achieved a **4×6 client matrix deployment** — all CLs onboard, with **Erigon and Reth** onboarded on the EL side.
- A **pending builder deposit** was in flight, expected to process at the end of the current epoch, enabling builder-produced blocks before the close of the ceremony.
- Considered the headline achievement of the week: a stable-ish ePBS devnet *and* BAL devnet running concurrently.

## 3. Block-Level Access Lists & Optimization Results (Toni)

### Status
- Dashboard numbers at presentation time were still based on **bal-devnet-2**; bal-devnet-3 results in progress.
- Benchmarks scope: parallel execution and batch pre-fetching. **Parallel state-root computation not yet tested.**

### Findings
- Worst-case benchmarks (e.g. maxed SLOADs, pre-fetcher adversarial cases) already show **significant performance gains** after only a few days of optimization work.
- **Reth** showed strong early results; **Besu** showed consistent behavior across the test set.
- Confirmed empirically:
- Parallel execution works.
- Batch pre-fetching delivers the expected efficiency gains.

Final numbers pending completion of the bal-devnet-3 benchmark runs.

**Acknowledgements:** Luis, Maria, Carlos, Jochem, Rafael for benchmarking work.

## 4. Repricing Results (Maria)

### EIP-7904 (Compute / Opcode & Precompile Repricing)
- Run against bal-devnet-3 with optimizations enabled.
- Many opcodes initially flagged for repricing **no longer require it** post-optimization.
- Remaining increases are **mostly small**.
- Largest required increase: **`POINT_EVALUATION` precompile** (expected).
- Other affected categories: a few precompiles and some arithmetic ops, all small.
- **Backward-compatibility impact significantly reduced** vs. the initial analysis (which was breaking many contracts). Re-run of the backward-compat analysis with new numbers still required.

### EIP-8038 (State Access Repricing)
- Run on "Jochem-net" — mainnet-sized snapshot plus a custom test contract.
- Anchor target: **100 Mgas/s**.
- Pre-optimization required increases were ~50–80×; post-optimization they are dramatically lower but still significant.
- Worst-case client driving the numbers: **Erigon**.

| Parameter | Required increase (worst client) | Notes |
|---|---|---|
| Storage access (10 GB contract, ENS-sized) | Slight increase | Acceptable |
| Account access | >2× | Reth is worst case; if Reth improves → ~1.9× |
| Account writes | ~4× | Improvement needed in Besu and Reth |

- Further reductions will require **across-the-board** client improvements rather than single-client wins.
- **Trade-off lever:** the 100 Mgas/s anchor is aggressive. Relaxing it would allow smaller cost increases at the price of a **larger block witness**.

**Acknowledgements:** Luis, Rafael, Jochem, Carlos for the benchmarking effort.

## 5. Gas Limit Target Decision

### Discussion summary
- Goal: agree on a **conservative public floor** for the post-Glamsterdam gas limit — a number the team is confident in delivering.
- Rationale: under-promising and over-delivering is preferred; a missed promise (e.g. claiming 5× and shipping 2×) damages credibility.
- Initial proposals ranged from 150M (most conservative) to 250M.
- Parithosh recalled the original theoretical math: 60 Mgas/s/thread × ~3–4 threads × 5s execution window (with ePBS) ≈ **900 Mgas theoretical ceiling**. 300M considered a stretch goal; **200M considered very realistic**.

### Bandwidth considerations
- Two EIPs reprice data costs, so bandwidth/block size will not increase materially.
- Toni: at 300M gas, worst-case block size remains roughly equivalent to today's.

### Open concerns flagged
- **Deposit throughput vs. SSZ limit:** at 200M gas, EL→CL deposits could exceed the 8,192 SSZ list limit. Mitigation: **Progressive Containers**.
- **State growth:** noted as no longer a blocker — addressed by the work done this week. Several participants observed that *if state growth is truly resolved, the realistic ceiling is meaningfully higher than 200M.*

### Decision
**Public floor target: 200M gas post-Glamsterdam**, communicated as a conservative minimum rather than a ceiling.

Rollout philosophy:
- Validate progressively on devnets, then testnets (with spamoor-style load), then mainnet.
- Not a single-step jump from 60M → 200M.
- Communicate to the community that Glamsterdam delivers a **multi-x scaling step**, abstracting away the underlying ePBS/BAL/repricing complexity.

## 6. FOCIL (EIP-7805) — Stretch Goal Update (Barnabas)

- No Dora page yet.
- **Three CL teams** working on implementations this week: **Lodestar, Lighthouse, Teku**.
- CL implementation based on the latest EIP-7805 spec, on top of the current devnet branch.
- EL side built on top of bal-devnet-6 branches.
- Goal of reaching the implementation threshold by end of the week was met.
- More EL/CL coverage expected in the coming weeks.

## 7. Awards & Acknowledgements

**Organizational / non-code prizes:**
- **La Donna** — overall organization
- **Louis and Josh** — filming the week

**Team prize:**
- **EthPandaOps** team

**MVP awards (Barnabas presenting):**
- **Maria**
- **The entire Erigon team** — recognized for sustained late-night effort despite staying off-site

**Best whip of the week:** tied — competition deferred.

## 8. Closing Remarks

Tim Beiko announced this as his **last day in L1 R&D**, while continuing to work on Ethereum in other capacities. He thanked participants for the week and for the past several years of collaboration.

## Action Items

| # | Item | Owner(s) |
|---|---|---|
| 1 | Complete bal-devnet-3 benchmark runs and update BAL dashboard with final numbers | Toni / EthPandaOps / benchmarking team |
| 2 | Publish final EIP-7904 repricing numbers; rerun backward-compatibility analysis | Maria |
| 3 | Drive Reth and Besu improvements on account access and account writes to relax EIP-8038 increases | Reth, Besu teams |
| 4 | Test parallel state-root computation under BAL | Client teams |
| 5 | Confirm builder-produced block on the ePBS devnet via the pending builder deposit | EthPandaOps / ePBS implementers |
| 6 | Investigate Progressive Containers as mitigation for the 8,192 SSZ deposit limit at higher gas targets | Spec / CL teams |
| 7 | Communicate **200M gas** as the conservative public post-Glamsterdam target | Tim / comms |
| 8 | Expand FOCIL (EIP-7805) implementation coverage beyond Lodestar, Lighthouse, Teku in the coming weeks | CL teams |
| 9 | Plan progressive testnet ramp using spamoor-style load before any mainnet gas-limit move | EthPandaOps |