Author here.

These benchmarks were run locally and published with full raw JSON so others can reproduce or challenge the results under identical conditions. The engine is focused on deterministic incremental computation rather than throughput-at-all-costs.

Happy to answer concrete technical questions about the test setup or metrics.