Context

Buildernet nodes rely on a software sidecar to receive orderflow and share it to other peers. This sidecar is called https://github.com/BuilderNet/FlowProxy, and has been developed by us with the goal of reducing the propagation latency of such orders and increase the observability of the system.

One of the observability tools introduced is an indexer of bundles and bundle receipts, which saves to a centralized Clickhouse server. A bundle receipt measures the time between a node receiving for the first time a bundle from the eth_sendBundle endpoint, and a peer receiving via the “system” family of sharing endpoints.

More precisely, a bundle receipt consists of the following fields:

-- DDL to create a ClickHouse table for storing bundle receipts data.
CREATE TABLE bundle_receipts (
  `bundle_hash` FixedString(32),
  `double_bundle_hash` FixedString(32),
  `sent_at` Nullable(DateTime64(6, 'UTC')) COMMENT 'The time the bundle has been sent at, as present in the JSON-RPC header',
  `received_at` DateTime64(6, 'UTC') COMMENT 'The time the local operator has received the payload',
  `dst_builder_name` LowCardinality(String) COMMENT 'The operator that received the payload',
  `src_builder_name` LowCardinality(String) COMMENT 'The name of the operator which sent us the payload',
  `payload_size` UInt32 COMMENT 'The size of the uncompressed payload in bytes',
  `priority` UInt8,
)

In this document we show some of the insights we can get from analysing this data over time.

Methodology

We will be examining the following quantities:

• signer distribution between different links. We will find out who are the most popular signers, uncovering region preferences and possible spamming activity.
• latency and bundle loss across different links and portions of the slot. Those two express the reliability of the system during congestion periods, and differentiating by link better highlights which challenges may happen during inter-continent communication.

We will also filter by individual continents (Europe, US) to uncover possible patterns that might be lost when looking at global data.

Time range examined

We’ll examine a day of data from 2025-11-06 00:00:00 UTC to 2025-11-06 23:59:59 UTC. From the Grafana dashboard, this range shows some outliers in high RPC call requests duration and HTTP open connections from the buildernet_beaver_azure_eastus_07 to other peers in west Europe. We can also observe a spike in HTTP failure both inter-continent and intra-continent.

A Grafana view of RPC call duration and HTTP connections/failures during 2025-11-06.

Signer distribution

The table below shows a leaderboard of signers, ordered by the number of unique bundles (distinct bundle hash) that have been sent to BuilderNet.

Leaderboard of signers, by number of unique bundles sent. On first place, we find SCP signer, while in second place there is Beaver pool.

From the leaderboard we can assess there isn’t a real spamming vector currently in the network, however there are two addresses with a very high duplicate percentage: