What happened

A Leios dozen-devnet benchmark compared two runs separated by a mempool configuration choice: timeouts enabled to bound the mempool tail, versus an unbounded configuration. In the bounded run, throughput increased from 66.6 to 88.3 to 96.6 kB/s as generators were added, reaching about 1.93 times the run’s 50 TxkB/s target.

The same record reports certification improving from 78% to 99%. “TooLate” votes fell from 78 to zero. By contrast, the otherwise-identical unbounded run is described as collapsing as load increased.

Why it matters

The paired result points to consensus vote timeliness as the relevant constraint in this test, rather than mempool ingest capacity alone. The supplied evidence says ingest scaled in paired runs, but on-chain throughput diverged according to whether votes reached their certification window.

For node and protocol engineers, the result identifies a concrete tuning and design concern: critical-path work should remain constant or logarithmic in mempool depth. The bounded setting is presented as a configuration choice, not a protocol change.

What to watch next

Watch further Leios dev-track results for certification rate, late-vote behavior, and comparisons between bounded and unbounded mempool configurations. Mempool depth alone may not describe the load constraint highlighted by this benchmark.

The useful next receipt would be additional devnet runs that test whether the same relationship holds beyond this paired comparison, before treating it as guidance for any mainnet rollout.

What to watch

Further devnet benchmarks showing certification, late-vote behavior, and throughput under bounded versus unbounded mempool configurations.

Sources and limits

Upstream references and independent checks

Digest dated 2026-08-23 · upstream model claude-sonnet-4-6. Direct links are available for 2 of 3 upstream source IDs; unresolved IDs remain visible for audit.

  1. 1
    dozen-devnet: the bounded run, and the tail is the whole storyDirect upstream source · 4705fc8d1e879637f3a3404d47107756043688fc
  2. 2
    dozen-devnet: record the phased unbounded run, and why it collapsesDirect upstream source · efc9e878a8cb0ccd78d16bc70997d1c08f1b9c8a
  3. 3
    04af1aa5846bbb4c3ade303de7692bb9aca01066Upstream reference; direct URL unavailable.

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: This is a thin, devnet-only record drawn from a single Leios pull request and one bounded-versus-unbounded run pair; no mainnet parameter has changed.