Wait for TestFlight before Claude build analysis

[mokagio]

Description

The :claude: 🕵️ Check for Build Failures step lists three groups in depends_on and testflight_builds_group is not one of them. It therefore fires as soon as the linters, prototype builds and unit tests settle, and upload-claude-analysis.sh snapshots the build's job states at that instant — where a job still running counts as not-failed. The TestFlight uploads are the longest jobs in the pipeline, so on trunk they are almost always still running, and they are the only jobs that can detect a release-pipeline outage.

That is not hypothetical. Trunk's TestFlight uploads had been rejected by App Store Connect with error 90186 since the 27.1 train closed, and the analyser reported it away twice on 2026-08-17:

Build	Uploads failed	Check ran	Result
#33780	04:32:24Z	04:31:31Z	saw zero failures, skipped the analysis entirely — no annotation
#33786	12:48:50Z	annotation published 12:48:45Z	"The build has one real failure: the Unit Tests job. Everything else either passed or is in expected states."

#33753 diagnosed the outage correctly on 08-13 — naming 90186 and recommending a pre-flight version check — only because an upload happened to fail 28 seconds before the check fired. That 28 seconds is the whole difference between the one correct report and twelve days of silence.

Two things a reviewer should know:

• This closes the instance, not the class. The analysis still asserts a global all-clear over whatever snapshot it is handed. Making it count non-terminal jobs and say "N jobs still running, not covered" is the general fix, and is deliberately not in this PR.
• The delay is trunk-only. testflight_builds_group is gated on build.branch == 'trunk', so PR builds skip it and the new dependency resolves immediately. On trunk, analysis now waits for the uploads — which is the point.

Testing instructions

Nothing to test in the app; this is a pipeline-graph change and only Buildkite can exercise it.

The behaviour it relies on is already demonstrated in the builds above: prototype_builds_group is PR-only, so on trunk builds #33780 and #33786 its jobs are broken and the check step ran anyway, because of the allow_dependency_failure: true already on the step. testflight_builds_group resolves the same way in reverse on PR builds.

To confirm on this PR: the check step should still run and reach its "All steps passed, skipping Claude analysis" branch despite both TestFlight and prototype dependencies being skipped. The real confirmation is the next red trunk build, where the analysis should now start after the uploads finish rather than before.

[dangermattic]

	1 Message
📖	This PR is still a Draft: some checks will be skipped.

Generated by 🚫 Danger

[wpmobilebot]

📲 You can test the changes from this Pull Request in WordPress by scanning the QR code below to install the corresponding build.

	App Name	WordPress
	Configuration	Release-Alpha
	Build Number	33804
	Version	PR #25905
	Bundle ID	org.wordpress.alpha
	Commit	4f4bd6f
	Installation URL	1g2bd1ljffi5o

Automatticians: You can use our internal self-serve MC tool to give yourself access to those builds if needed.

[wpmobilebot]

📲 You can test the changes from this Pull Request in Jetpack by scanning the QR code below to install the corresponding build.

	App Name	Jetpack
	Configuration	Release-Alpha
	Build Number	33804
	Version	PR #25905
	Bundle ID	com.jetpack.alpha
	Commit	4f4bd6f
	Installation URL	0vi95mmrudt3g

Automatticians: You can use our internal self-serve MC tool to give yourself access to those builds if needed.