Airflow, Dagster, Prefect: figures withdrawn
This page carried scheduler numbers that came from the site's own build fixtures. They were never measured, and they are gone.
What was here
A table of scheduler latencies for Airflow, Dagster and Prefect, a headline figure of
p50 340 ms → 180 ms, a claim of 200 runs each, and a link to a harness at
github.com/example/dep-harness-orchestrators.
None of it was measured. The numbers were written as fixtures while this site was being built — they existed so that the article template, the feed card and the correction machinery had something realistic to render — and they were published by mistake when the site went live.
Why the page still exists
Because this publication’s rule is that an address keeps resolving and an article is corrected rather than withdrawn. Deleting the page would be the more comfortable option and the less honest one: anyone who linked to it, or read those numbers and repeated them, would find nothing here and no explanation.
So the address stays, the correction above records exactly what changed, and the numbers are gone. If you took a figure from this page, it was not a real one.
What is actually coming
The comparison is worth running and it is queued. Two questions from it are in the plan: what each orchestrator costs while idle — which nobody appears to have published — and what the Airflow 2 to 3 upgrade does to scheduler performance, where the migration is well documented and the performance consequence is not.
When either lands it will carry a harness you can run and the raw rows behind every number, because that is the only thing that distinguishes it from what was on this page.