Eighteen workflows have never succeeded — 8182 runs, zero green

[gHashTag]

Eighteen active workflows across three repositories have never succeeded
once. Between them they have consumed 8182 runs and produced zero green
results.

repo	workflow	runs	successes
trinity	S³AI Brain CI — "Functional MRI" Gate	1705	0
trinity-fpga	queen-bot · PR auto-merger	1541	0
t27	auto-merge-ready-prs.yml	1224	0
trinity	Agent Complete — Destroy + Drain	565	0
trinity	KOSCHEI Production Deploy	475	0
trinity	docs-check	454	0
trinity-fpga	FPGA CI	330	0
trinity-fpga	🧠 FPGA Consciousness Regression	328	0
trinity	CI Runner	251	0
trinity	Spec Audit	247	0
trinity	Claude Code	238	0
trinity	Trinity Agent Decompose	207	0
t27	Coq kernel	166	0
trinity	Terminal Demo GIFs	163	0
t27	OpenSSF Scorecard	125	0
trinity	🧠 FPGA Consciousness Regression	57	0
trinity	Trinity Nexus Build	56	0
t27	Coq Proofs Validation	50	0

Measured with, for each active workflow:

gh api "repos/OWNER/REPO/actions/workflows/<id>/runs?per_page=1" --jq .total_count
gh api "repos/OWNER/REPO/actions/workflows/<id>/runs?status=success&per_page=1" --jq .total_count

Only workflows with at least 50 lifetime runs are listed, so none of these is
new or rarely triggered.

Why this matters more than any one of them

A gate that has never been green carries no information: it is red before your
change and red after it, so nobody reads it, and after a while nobody reads the
others either. That is not a hypothesis here — it is the measured cause of
the nine defects found in verify-request.yml this week, in a path that had
run once in its life and produced a report telling a stranger their clean
design had a silicon bug.

Eight thousand red runs is also the training set for ignoring red runs.

What to decide, per workflow

Three outcomes, and each is fine:

1. Fix it — it guards something real and is simply broken.
2. workflow_dispatch only — it is useful occasionally and should not run
   on every push.
3. Delete it — it guards nothing anybody acts on.

What is not fine is the current state, where the answer is "leave it red" and
the cost is paid by every other gate in the repository.

I am not making that call unilaterally across eighteen workflows in three
repositories; it needs the owner. This issue is the measurement.