Benchmarks in this industry are mostly untrustworthy, and usually for the same reason: the person running them chose the shape of the work. So let us be precise about what this is and what it is not.
This is not a comparison against dbt. dbt is not an execution engine — it generates SQL and hands it to your warehouse — so "Interlace versus dbt" would be measuring DuckDB against whatever you pointed dbt at, dressed up as a tool comparison. That number would tell you nothing.
This is a capability demonstration: a real DAG doing real work, on hardware you can buy, with every command written down. The point is not that a number is small. The point is that you can run the same thing in the next ten minutes and get your own.
The shape of the work
The examples/benchmark project generates 25 million synthetic events in-engine — nothing to download, fully deterministic — and pushes them through a deliberately awkward fan-out:
