What happened

A dozen-node Ouroboros Leios development-network benchmark increased transaction injection across three runs. Ingest rose from 78 to 156 to 197 kB/s, while confirmed on-chain throughput moved the other way, from 77.6 to 64.1 to 51.8 kB/s.

The associated commits attribute the decline to consensus votes arriving after the window needed to certify endorser blocks. The count of votes marked "tooLate" rose from 0 to 12 to 60 across the runs.

Why it matters

The result suggests that, in this test, accepting more transactions was not the limiting step. The critical issue was whether votes arrived in time on the consensus path, which makes certification rate under load more informative than mempool depth alone.

For node and protocol work, the stated design lesson is to keep critical-path work constant or logarithmic in mempool depth. The record says linear work on that path can reverse the expected relationship between higher load and higher confirmed throughput.

What to watch next

Watch for further devnet results that report certification rate and vote timeliness under load, especially comparisons between bounded and unbounded mempool configurations. Those results would clarify whether tuning can preserve certification windows as ingest rises.

This is a development-track signal, not a throughput target or a released Cardano protocol change. It does not establish a mainnet outcome.

What to watch

The next useful receipt is a comparable bounded-versus-unbounded devnet run showing certification rate, late-vote counts, ingest, and confirmed on-chain throughput under load.

Sources and limits

Upstream references and independent checks

Digest dated 2026-08-23 · upstream model claude-sonnet-4-6. Direct links are matched to all 3 upstream source IDs.

  1. 1
    dozen-devnet: record the phased unbounded run, and why it collapsesDirect upstream source · efc9e878a8cb0ccd78d16bc70997d1c08f1b9c8a
  2. 2
    The limit ends up in consensus, not mempool or tx-submissionDirect upstream source · 13868a10c7a4f4eddda96b85298c3f2640e1c3d0
  3. 3
    dozen-devnet: the bounded run, and the tail is the whole storyDirect upstream source · 4705fc8d1e879637f3a3404d47107756043688fc

This Research brief was generated by Terra from a dated upstream research digest. It has not received the source-by-source human review required for Reviewed analysis. Material limit: The evidence is limited to a private dozen-node devnet benchmark and commits from one Leios pull request by a single author; it is not mainnet evidence, a released change, or a Cardano throughput forecast.