Reporting window: September 21, 2026, 00:00:00 to September 28, 2026, 00:00:00 WITA (Asia/Makassar), end exclusive. This corrected edition covers the complete ISO week 39.

Ethereum developers have committed Sepolia to signaling a 200-million gas limit at Glamsterdam, despite reports of slow valid-block execution. Meanwhile, Nixo’s merged Hegotá scoping update records ten proposals moving to Considered for Inclusion. The editorial priority is the tension between those tracks: future scope is becoming clearer, while the imminent testnet transition still needs convincing performance evidence. The next decisive signal should be how clients handle adversarial blocks, beyond whether the scheduled fork activates.

Sepolia will test the higher limit at the fork

Justin Traglia merged the gas-limit schedule on September 24 following ACDE #246. His proposal specifies immediate signaling, but the actual limit takes hours to reach the target. That distinction matters when assessing the transition: the schedule does not put a 200-million-gas block at the fork boundary by fiat.

The decision followed a concrete warning. On the call, potuz reported valid JUMPDEST-analysis workloads taking more than eight seconds. Some client reports reached 13–14 seconds. Ansgar Dietrichs argued for demanding testnet conditions to expose problems before mainnet. These are attributed call reports, not a standardized cross-client benchmark.

The testing response has begun. On September 25, execution-specs merged chfast’s randomized JUMPDEST-analysis benchmark. The benchmark pre-funds CREATE destinations and varies initcode to expose branch-prediction costs. Watch comparable reruns and client fixes against that workload. The benchmark merge establishes better test coverage, not resolution of the reported slowdown.

Compiler support advances Hegotá’s subroutine proposal

EIP-7979: Call and Return Opcodes for the EVM moved to CFI in the September 24 decision record. Solidity’s Radek supplied an important qualification: the team’s earlier optimization objections were addressed, and Solidity would use the instructions if accepted. He described the compiler benefits as fairly minor. Besu’s Justin Florentine then said Solidity’s input made Besu open to raising its rating from B to A. The meaningful change is implementer support, without treating the proposal’s performance ambitions as measured network gains.

Developers kept a separate question open. EIP-8368: CPSB Recalibration for New Gas Limit and EIP-8372: Normalized state gas limit remain PFI pending 30 days of mainnet Glamsterdam data. The decision record allows considering one proposal or declining both. That evidence requirement is the next signal to watch, rather than assuming CFI means the broader gas-accounting design has converged.

Hegotá also gained a new implementation venue. On September 24, Stefan Starflinger reported that the Frames devnet had launched the previous day. The report marks the start of devnet work, not a passed interoperability test. Watch participating clients and reproducible test results.

Gloas fixes head selection while Quick Slots remains experimental

On September 24, jihoonsong merged 0xsamalt’s payload-status viability correction. The PR identifies a Gloas path where a non-viable payload-status variant could survive filtering and become the selected head. This is a fork-choice specification correction. Client adoption and regression results remain the evidence needed to establish deployment readiness.

On September 25 WITA, consensus-specs merged the feature specification for EIP-8198: Quick Slots. Parameter convergence remains unfinished: the PR describes an intended first reduction to ten seconds, with no activation epoch, while the published EIP still specifies eight seconds. The Hegotá meta-EIP still lists Quick Slots as PFI. Treat the merge as infrastructure for review and prototyping. Watch agreement between the documents and tested timing budgets before reading a shorter-slot commitment into the merge.

Payload segmentation gains a second simulator check

Csaba Király’s September 23 research post retested segmented payload diffusion in Shadow over QUIC and TCP. His preferred design reproduced the earlier harness results within a few percent, while queue modeling changed the baseline and the ranking of alternatives. That cross-check strengthens the research case without establishing one universal speedup.

The boundary is explicit: the tests omit a full client and do not validate the bandwidth model or per-node CPU cost. The separate Linux check used 100 processes on one machine. The useful next step is full-client evidence under realistic contention, especially alongside the execution delays discussed at ACDE.

Sources and further reading