Research

Experiments we ran on real open-source systems. Each write-up gives the recorded numbers and how the experiment was run.

  1. listmonk said the campaign finished. 24 of 40 subscribers never got it.

    An agent with only the source code and our docs asked whether a restart in the middle of a campaign loses any sends. One recorded run says it can, and listmonk’s own query shows why.

    listmonk v6.2.0, 14 November 2026

  2. NATS JetStream under the same faults, with and without a sync on every write.

    A known issue, first published by Jepsen, reproduced as a side-by-side comparison: 31 of 100 runs lost acknowledged writes with the default settings, and none did with sync_interval: always.

    NATS JetStream 2.12.1, 14 November 2026

How we report results

Every number comes from a recorded run.

Each write-up says what was tested and how many times. If a figure is an illustration or a model, the page says so next to it. When we reproduce something others found first, we link to them and say so at the top.

Have a system you would like looked at this way?

Request early access All benchmarks