Four workflows are red on master today, and two of them are green by not running

[gHashTag]

Nine workflows had a failure as their most recent master run, all of them from one mass dispatch on 2026-08-28T18:32–19:28Z, and none had been re-run since the fix PR #2763 landed at 20:00Z. Their red was a five-day-old reading nobody had retaken.

I retook it for the five that have no outward effect. Dispatched on master, 2026-08-29T06:17Z:

workflow	verdict today	why
Check Now Freshness	success	#2763's NOT APPLICABLE path works
Build Paper	failure	Build PDF from LaTeX — and main.pdf was written (8 pages, 365 373 bytes), so the step fails after a successful compile
Coq Proofs Validation	failure	Install Coq Interval: "Opam has not been initialised" — no opam init before opam install
Lean proofs	failure	lake build; lake check-test failed, Cache not found for input keys: lake-Linux-X64-…
Vivado Synth (Docker)	failure	Initialize containers: Error response from daemon: manifest unknown — the image it pulls does not exist

Four are red on master right now. Not decayed readings — reproducible today.

Why nothing reports it

build-paper     push: branches=[main, master]   paths: YES   -> green by not running
coq-proofs      push: branches=<all>            paths: YES   -> green by not running
lean-proofs     push: NONE                                   -> never runs on master
vivado-synth    push: NONE                                   -> never runs on master

Two are the paths:-filter hazard class from #2754 (16 of 46 workflow files fire on push-to-master and carry a paths filter). Two are dispatch-only — lean-proofs deliberately so, per its own header comment: "get the reading, then decide whether to gate." This is that reading, and it is red, which is what #2747 predicted.

Vivado Synth looks like a dependency, not an independent defect: the image it pulls is built by Build Vivado Docker Image, which is also in the failing nine.

Not dispatched, and why

Four of the nine have outward effects and are the owner's call, not mine:

workflow	effect
Release Pipeline	5 publish steps
Deploy Control Plane API	deploys
Brain Seal Refresh	git push
Build Vivado Docker Image	docker push

Their last-run verdicts are still the stale 2026-08-28 readings. Someone with the authority to run publishing workflows should retake them.

Method note

My own first reading of master health in this session was gh run list --branch master --limit 60, which returned all-green. That window spanned 2 hours 28 minutes. Widened to 1 000 runs over five days: 940 success / 60 failure across 40 distinct workflows. A run-list limit is a time window in disguise.

Refs #2762, #2754, #2747