[Outcome Report] Workflow Health Report — 2026-06-04

[github-actions]

Workflow Health — 2026-06-04

Executive read: 10 pending items concentrated in Smoke Claude and Changeset Generator; most outcomes are commented/reviewed (medium evidence). Acceptance rate at 78% is solid, but zero-touch rate of 0% indicates all outputs required human engagement. Several items without trackable URLs (pending with "no url" detail).

Workflow	Status	Lifecycle health	References
Smoke Claude	🟨🟨🟥🟩🟨🟨🟨⬜🟩⬜🟨🟨	🟡 in flight	🟨 36748 · 🟥 36748 · 🟩 36748 · ⬜ 36773 · ⬜ 36766 · 🟨 36748
Smoke Gemini	🟨🟩⬜🟩🟩	🟢 resolving	🟨 36774 · 🟩 36769 · ⬜ 36770 · 🟩 36748 · 🟩 36748
Smoke Codex	⬜🟨	🟡 in flight	⬜ 36771 · 🟨 36748
Changeset Generator	🟨🟥	🟡 in flight	🟨 36769 · 🟥 36769
Agent Container Smoke Test	🟩🟩	🟢 resolving	🟩 36769 · 🟩 36748

Legend:

• Status: 🟩 accepted · 🟥 rejected · 🟨 pending · ⬜ unknown
• Lifecycle health: 🟢 resolving · 🟡 in flight · 🟠 aging · 🔴 stuck · ⚪ underdefined
• References: one linked item per status emoji, in the same order as the Status column

🔴 Action Items

1. Smoke Claude — 7 pending items + 2 unknown — This workflow has the highest pending volume. Review the pending create_pull_request_review_comment, post_slack_message, create_code_scanning_alert, and create_check_run items marked "no url" — these cannot be evaluated without proper tracking URLs. Need to verify if these tools succeeded but failed to return URLs, or if they failed silently.

2. Items without trackable URLs — 5 outcomes have detail: "no url" (Smoke Claude: 4 items; untracked). These are all in pending state and cannot be verified. Either the safe-output tools are not returning URLs, or the workflow is not capturing them. Needs investigation.

3. Zero-touch rate at 0% — All 7 accepted items required human engagement (comments, reviews, edits). This suggests the agents' outputs are always incomplete or require human review by design. Flag: Is this expected, or should agents produce more self-contained outputs?

4. Acceptance rate at 78% — Solid but below ideal. The 2 rejected items (update_pull_request in Changeset Generator, add_reviewer in Smoke Claude) should be reviewed to understand why outputs were not retained.

Detailed metrics, evidence quality, workflow counts, and trends

Outcome Scorecard — 2026-06-04

Metric	Value	Status
Acceptance rate	77.8%	🟡 60-80%
Zero-touch rate	0%	🔴 <25%
Waste rate	8.7%	🟢 <10%
Median time to resolution	6 minutes 27 seconds	—
Accepted	7 / 23	—
— strong evidence	0	(none)
— medium evidence	7	acted on, state retained/replaced
— weak evidence	0	(none)
Rejected	2	—
Ignored	0	—
Zero-touch	0 / 7	—
Pending	10	—
Runs checked	7	—

Per-Workflow Breakdown

Workflow	Accepted	Rejected	Ignored	Pending	Unknown	Total	Acceptance	Waste
Smoke Claude	2	1	0	7	2	12	66.7%	8.3%
Changeset Generator	0	1	0	1	0	2	0%	50.0%
Smoke Codex	0	0	0	1	1	2	—	0%
Smoke Gemini	3	0	0	1	1	5	100%	0%
Agent Container Smoke Test	2	0	0	0	0	2	100%	0%

Sort note: Sorted by waste rate descending (Changeset Generator worst first).

Evidence Quality

No items evaluated with weak existence-only signals (fallback_exists_only_count = 0). All accepted items have medium evidence (acted on, state retained). No strong evidence outcomes (merged PRs, closed issues, completed tasks with state verification).

Tracking & Data Quality Issues

⚠️ 5 items with missing URLs — These outcomes are in pending state but have url: "" and detail: "no url":

• create_pull_request_review_comment (Smoke Claude, 2x)
• post_slack_message (Smoke Claude)
• create_code_scanning_alert (Smoke Claude)
• create_check_run (Smoke Claude)
• add_labels (Smoke Claude)

These cannot be verified as accepted or rejected. Recommendation: Check if the safe-output tools are designed to return URLs for all output types, or if these tool types should not generate trackable artifacts.

Trend Comparison (vs. 2026-06-01)

Previous (2026-06-01):

• Acceptance rate: 90.9%
• Waste rate: 3.7%
• Zero-touch rate: 0%

Current (2026-06-04):

• Acceptance rate: 77.8% (⬇️ down 13.1pp)
• Waste rate: 8.7% (⬇️ up 5.0pp)
• Zero-touch rate: 0% (➡️ stable)

⚠️ Regressing: Acceptance rate dropped significantly (13pp), and waste rate increased (5pp). This suggests workflows are producing less polished outputs or targeting less amenable items. Investigate: Did the prompt change, or did the sample shift to harder problems?

📊 Measured by Outcome Collector · haiku45 41.3K

• expires on Jun 11, 2026, 1:09 AM UTC