Read-only PR review timing brief for Airflow: long-tail review waits

[xpinc17]

Hi Airflow maintainers,

I put together a short read-only PR review timing brief for apache/airflow, based only on public GitHub data. Nothing was installed, and nothing was posted or changed in the repo.

This is not meant as a criticism of the project. Airflow has a very active PR queue, so I was curious whether a lightweight historical read could surface any useful long-tail review patterns.

What stood out

For PRs created between 2026-05-29 and 2026-06-08:

Metric	Observed value
PRs analyzed	458
PRs waiting >48h for first review	29
Approved PRs idle >48h after last approval	19
Total review-wait, cumulative across PRs	101.2 PR-days
Median first-review wait	2.2h

The median first-review wait looks healthy, but there is still a visible long tail: some PRs waited several days for first review, and some approved PRs remained unmerged for more than 48h.

Longest observed first-review waits

These are longest observed waits in this window, not typical PRs:

• #67675 — 7.6 days
• #67724 — 154.9h
• #67730 — 125.4h

Why I’m sharing

The main pattern is not “reviews are slow” overall. It is more specific: in a very active repo, aging review items can fall out of view even when the typical PR moves quickly.

ApprovalOps is a small tool I’m building to surface those aging review items earlier. In this historical window, it would have flagged an estimated 40% of observable review-wait time earlier, about 40.1 cumulative PR-days. That is a retrospective estimate, not a guarantee.

Caveats

• This is based only on public GitHub data.
• It does not know internal maintainer context, reviewer load, branch protection, or project policy.
• Aggregate first-review metrics may use merge or close time as a fallback when no first-review timestamp is visible.
• The examples above only include PRs with observed first-review timestamps.

If this is useful, I’m happy to share the methodology or run the same read-only brief for a different timeframe.

[potiuk]

You can take a look at the stats we are gathering with PR triage - coming from https://github.com/apache/airflow-steward. which we are going to use very soon to improve our review procesess https://lists.apache.org/thread/80lmskw22k10clw4nh4j0yvsrsjl0cpo

If you have any improvements to the triage process and magpie-pr-stats SKILL that we have there - your contributions to "airflow-steward" (soon hopefully "Apache Magpie" Top Level Project of ASF) are most welcome.