Post-Quantum (PQ) Interop #48, July 22, 2026

Protocol Calls & happenings
1

Agenda

  • devnet issues
  • flock/leanVM-b

Meeting Time: Wednesday, July 22, 2026 at 14:00 UTC (60 minutes)

GitHub Issue

2

Meeting Summary:

This meeting was a client update session where teams discussed progress on bug fixes, optimizations, and performance improvements for their respective clients in the DevNet environment. Parthasarathy from the Zoom team reported fixing bugs discovered in recent runs but noted that justification and finalization were stalling at around 12,000 and 9,000 slots respectively due to slow P95 times of 4 seconds for block production and aggregation. Shariq from the ETH Lambda team achieved much better results with 44,941 slots of finalization on a 4-subnet DevNet with 6 clients per subnet, though they experienced out-of-memory issues with Veeam nodes. The ETH Lambda team also implemented ETH P2P as an additional transport layer for interop with ETH Lambda. Other teams including Lantern, Gene, and Grindage reported focusing on stability improvements and bug fixes, while Thomas provided an update on research work involving binary field techniques and proving systems, noting current performance was about 4x slower than the classical prime field version. The discussion included questions about which optimization proposals had been implemented across different client teams, with some overlap between ETH Lambda and Zim proposals being clarified.

Click to expand detailed summary

Parthasarathy reported that the Zoom team spent the past week fixing bugs discovered in both their run and the DevNet run by Eve Lambda. The team identified performance issues where justification stalls after 12,000 slots and finalization stalls around 9,000 slots, primarily due to P95 delays of 4 seconds for block production and aggregation. To address these issues, the team implemented ETH P2P and conducted interop testing with ETH Lambda as an additional transport layer, which is expected to be fully replaced by EEC-P2P once EEC-Lambda is ready.

Shariq reported running a DevNet with 4 subnets and 6 clients per subnet, achieving 44,946 slots before Veeam nodes crashed due to out-of-memory issues. Parthasarathy mentioned running an 8-subnet DevNet with similar interoperability issues across different clients. Shariq also mentioned integrating with Shadow and running small simulations, suggesting collaboration with other teams to include Dream in their Shadow simulations.

Mihir reported that the team focused on ensuring stability of Lantern in devnets, implementing optimizations including block building and integral computation, and plans to add additional servers for multi-subnet devnets. Pablo updated on aggregation improvements with early start and recursive prioritization, HP2P compatibility work including ALPN stream model implementation, and an alternative execution client integration that embeds Etherex into EthLambda.

The team discussed improvements made to Gene on DevNet 5 multi-client runs, focusing on addressing issues that made it appear unhealthy during certain conditions. Key fixes included resolving a peer stalling issue and improving checkpoint sync behavior to retry when sources are temporarily unavailable. Boma reported optimizations made to their client and mentioned EPF fellows contributing to DB work, with plans to release a new image the following day containing optimizations and LeanSpec fixes.

Thomas provided an update on the research and development work, focusing on binary field techniques and the binary version of LinVM. He explained that while the original LinVM with Poseidon and Koala Bear is on hold, Emil is working on improving the binary field version to enhance performance. Thomas noted that while the current binary field proving system shows a 4x slowdown compared to the classical version, ongoing optimizations are in progress to address this issue. The team is also working on improving the BINU64 library to match or exceed the performance of Flock, including aggregation capabilities.

Mega presented Lambda’s work on tooling to measure latency of block proposals and attestation arrival times using an API ported from the Beacon API. The tool features a centralized collector that listens for events and timestamps them, with a frontend interface showing arrival times and color-coded aggregations. Thomas confirmed his understanding of the graph’s color coding and noted the significant variations in aggregation times shown in the data.

Mega and Thomas discussed issues with aggregation times in LeanVM, particularly noting slow aggregation causing delays in block production. They identified that the right side of the graph shows problematic aggregation times, likely due to regulatory aggregations. Thomas suggested exploring a smaller signature version of LeanVM that uses sub-MTU XMSS, which could potentially resolve the aggregation issues, though they noted this version currently lacks theoretical proof from cryptographers.

The team discussed implementation status and performance metrics of various optimization proposals across different client teams. Thomas and Mega noted they needed to verify benchmark numbers with Emil regarding speed improvements on specific branches. Shariq confirmed that ReamsHive implemented two ETH Lambda proposals: early bird optimization and removal of Type 1 proofs. Parthasarathy indicated that the Zim team had not implemented the slot time change proposal but had implemented other ETH Lambda suggestions. KaydenML mentioned they made fixes for tick interval metrics and the 128-slot reorg issue. The discussion revealed that different teams were at various stages of implementing the proposed optimizations, with some teams still needing to provide performance data.

The team discussed implementation gaps in the ZIM proposal regarding split-target votes, with Boma identifying concerns about the transition from specific source targets to shared source recognition that need safety arguments and implementation details. Parthasarathy clarified that while some overlapping proposals have been implemented, the 8-slot timing increase has not been implemented yet and will need verification. The group also discussed the ETH P2P implementation as a potential replacement for libP2P, with Parthasarathy explaining that ETH Lambda and Zim teams have implemented separate transport layers using both protocols, though performance comparisons are still pending. Thomas suggested reaching out to Camille from the EF for more information about ETH P2P developments.

Next Steps:

  • Thomas: Check with Emil about the possibility of switching to the LeanVM branch with smaller signatures to potentially improve aggregation performance, and post the question on the Telegram channel.
  • Parthasarathy: Verify the code to confirm which optimizations from the Zim proposal have been implemented by the Zoom team and report back.
  • Boma: Post details about the “split-target votes” proposal from Zim on the Telegram channel for clarification and discussion.
  • Shariq: Include other client teams in Shadow simulations and coordinate with them to include their nodes.
  • Parthasarathy: Once confirmation is received from all client teams about implemented optimizations, run a DevNet and share performance metrics.
  • Mihir: Reach out to Camille from the EF for more information about ETH P2P and potential performance improvements.
  • Boma: Release a new client image by tomorrow with optimizations and LeanSpec fixes.
  • All client teams: Continue running and scaling DevNets, and report any issues or blockers on the Telegram channel.

Recording Access:

3

YouTube recording available: https://youtu.be/qEUHMuTv0NM