Some quick data on mainnet attestation propagation. Noticeable improvements since @misilva73 's post from last year! ethresear.ch/t/an-analysis-o… → Once a block is released on the network, attesters cast a vote on it. These votes are then aggregated into "summary votes", and included in the following block. All of this (block release, attestation release, aggregation release) must fit within 12 seconds, today's block time. → The plot here measures the time that attestations are first seen by a network of sentries operated by @ethPandaOps, relative to the first-seen time of a newly released block, or 4s if the block was missed. → Orange curve (2026) is left of the blue curve (2025), meaning that attestation propagation time has broadly decreased on the network. This is great! → The curve shows some bimodality: two clusters of arrival times. 1. The first could be described as "normal behaviour": vote when the block is received, or vote when the 4s deadline hits before a block is received. 2. The second cluster mostly relates to nodes that vote at 4s even when they receive the block before the deadline. Some clients implemented this behaviour even if this was contrary to the specs. Some nodes may have been playing attestation timing games. → The good news: The first cluster used to cover ~50% of arrivals (high slope until a lil bend around 50%). Now the first cluster covers close to 80% arrivals! More nodes are more timely ⏰ → What explains it? More clients have implemented the right voting behaviour of releasing votes as soon as the block is received. → The good news 2: The first cluster got even faster! The fastest 50% of attestations in 2026 are faster than the fastest 50% of attestations in 2025. → What explains it? Performance improvements and a more consolidated validator set (close to ATH, 1.1 million validators during the 2025 analysis; now more than 10% smaller) likely explain most of the overall improvement. Overall this is tailwinds for faster slots. With our 10-second target for Hegotá, no improvement to attestation propagation could have still been good enough, given that ePBS in Glamsterdam pipelines the block propagation between beacon and execution, leaving more time for attestations and aggregations to propagate. That said, better attestation propagation benefits both future consensus and networking updates, such as decoupled consensus, as well as latency of the fast confirmation rule, on top of giving more breathing room for nodes located at the edges of the gossip network, ensuring geodecentralisation.