[portfolio] 📊 Portfolio Dashboard - 2026-04-06

[github-actions[bot]]

Overview

Analysis of 317 workflow runs over 3 days (Apr 4–6, 2026). Total spend: $51.39 (~$17.13/day). The portfolio has one dominant cost driver — Smoke Claude at $18.37 (36% of spend) — and a reliability concern with 38.8% of runs completing with errors. Most high-error workflows have zero token cost (smoke/integration tests), so wasted spend is concentrated in ~$6.21 of identifiable failures.

⚠️ Data Window: Only 3 days of history available. Patterns may not be representative of steady-state. Recommendations are conservative.

---

Visual Summary

Cost Trends (Apr 4–6)

Key Insights:

• Daily average: $17.13
• 3-day window only — insufficient for trend analysis
• Monthly projection: ~$514 if current pace holds

Top Cost Drivers

Top 3 workflows account for ~47% of total cost:

1. Smoke Claude — $18.37 (green: 14% error rate, triggered by PRs + schedule)
2. Go Logger Enhancement — $3.76 (red: 50% error rate)
3. Sergo - Serena Go Expert — $1.99 (green: 0% errors)

Failure Analysis

Wasted Spend on High-Error Workflows: ~$6.21

• 10+ workflows with >30% error rate
• Several at 100% failure (CI Cleaner, Smoke Gemini, Smoke Codex, AI Moderator)
• Most are zero-cost test workflows — reliability issue, not a cost issue

Overall Run Distribution

• Completed (no errors): 192 runs (60.6%)
• Completed (with errors): 123 runs (38.8%)
• In Progress: 2 runs (0.6%)

---

💰 Cost Reduction Opportunities

Total Identifiable Savings: ~$3.76–$9.50/month (7–18% reduction)

Strategy 1: Fix High-Error Costly Workflows — $3.76 wasted/3 days

These workflows have both significant cost AND high error rates:

Workflow	Error Rate	Runs	Total Cost	Wasted
Go Logger Enhancement	50%	2	$3.76	$1.88
Daily Documentation Updater	50%	2	$1.40	$0.70
Copilot Agent Prompt Clustering Analysis	50%	2	$0.88	$0.44
Step Name Alignment	50%	2	$0.82	$0.41

Fix: Investigate and fix root causes of failures. These workflows appear to be one-time runs or early-stage, so errors suggest broken implementations rather than systematic issues.

Potential savings: ~$3.43/period if error rates drop to ~0%

Strategy 2: Address High-Failure Zero-Cost Workflows — Reliability Risk

These workflows run frequently but fail constantly. While currently zero token-cost (they fail before incurring AI costs), they represent infrastructure waste and reliability risk:

Workflow	Error Rate	Runs (3d)	Notes
AI Moderator	96%	23	Schedule-triggered, near-total failure
Smoke Codex	95%	20	Codex engine failing consistently
Issue Monster	79%	19	High frequency, high failure
Changeset Generator	88%	17	Almost never succeeds
Smoke Gemini	100%	3	Complete engine failure
CI Cleaner	100%	3	Never succeeds
GitHub Remote MCP Authentication Test	100%	3	Auth failures

Fix: Disable or pause Smoke Gemini, Smoke Codex, CI Cleaner, and GitHub Remote MCP Authentication Test until underlying engine/auth issues are resolved. Reduce Issue Monster and AI Moderator schedule frequency while debugging.

Potential savings: Reduced CI minutes + faster failure detection

Strategy 3: Reduce Smoke Claude Scheduling — $9.50+/month

Smoke Claude is the single largest cost driver at $18.37 over 3 days. It runs on both PR events AND schedule:

• 3 scheduled runs in 3 days = ~1/day schedule
• 18 PR-triggered runs in 3 days

Recommendation: Review whether smoke tests need to run on EVERY PR vs. only on specific label or draft-to-ready transitions. If reduced from every PR to labeled PRs only, estimated savings: 30–50% of smoke test cost.

Potential savings: ~$3–9/month (conservative estimate based on 3-day window)

---

🎯 Priority Actions

1. CRITICAL — Disable Smoke Gemini, Smoke Codex, and GitHub Remote MCP Authentication Test (100% failure rate, running 3x/day)
2. HIGH — Investigate Go Logger Enhancement failure root cause ($1.88 wasted on 2 runs, 50% failure)
3. HIGH — Fix AI Moderator and Changeset Generator (96%/88% failure rates, very high run frequency)
4. MEDIUM — Add PR trigger filters to Smoke Claude to reduce unnecessary runs (top cost driver)
5. MEDIUM — Fix Daily Documentation Updater and Daily Copilot Token Usage Audit (high error rates on daily schedules)

---

📈 Data Quality

• Period Analyzed: 2026-04-04 to 2026-04-06 (3 days)
• Total Runs: 317 workflow runs
• Workflows executed: 70+ distinct workflows
• Confidence: Low-Medium — 3-day window limits trend detection; single-run workflows cannot show failure patterns
• Cost data: Derived from actual token usage via episode tracking

Methodology: Analysis based on pre-downloaded execution data from gh aw logs. Costs calculated from actual token usage tracked in episodes. Error rate based on error_count > 0 per run.

References:

• §24026181213 — Current workflow run

Generated by Automated Portfolio Analyst · ● 1.3M · ◷

• expires on Apr 7, 2026, 9:51 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-04-07T09:51:00.265Z.

Closed by Workflow