[api-consumption] 📊 GitHub API Consumption Report — 2026-06-09

[github-actions[bot]]

📊 GitHub API Consumption Report

Report Date: 2026-06-09 · Repository: github/gh-aw · Run: #27203760628

Scope note: This snapshot covers the current UTC day 2026-06-09 (00:00 → ~11:53 UTC) — a ~12h window, consistent with prior daily entries (which are also mid-day snapshots). REST API figures use github_rate_limit_usage.core_consumed (actual quota points consumed), read per-run from run_summary.json.

---

Today at a Glance

Metric	Value
🤖 Total Runs	187 (153 ✅ / 34 ❌)
🎯 Success Rate	81.8%
🔗 GitHub REST API Calls	207,880 (core quota consumed — includes reads, writes, and all GitHub API operations)
📝 Safe-Output Writes	89 (issues + PRs + comments + discussions created by safe-output tools)
⏱ Avg Duration	614.8s (p95: 1274.6s)

⚠️ Anomaly detected: today's 207,880 core-quota consumption is a record high — 2.6× yesterday (80,144) and 1.76× the previous peak (118,077 on 2026-05-27). Statistical z-score +3.64 against the 22-day history. See analysis below.

---

🔗 GitHub API Calls Trend (90 days)

Daily consumption sat in a stable 17K–84K band for most of the tracked window, with a single prior spike (~118K) on 2026-05-27. Today breaks sharply out of that range to 207.9K, lifting the 7-day rolling average to ~80K. The jump is concentrated rather than broad-based — see the per-workflow breakdown.

---

🔗 GitHub API Calls by Workflow Trend (30 days)

The four PR-gating reviewers — PR Code Quality Reviewer, Matt Pocock Skills Reviewer, Test Quality Sentinel, and Design Decision Gate — are consistently the heaviest consumers and move together day-to-day, since they all fire on PR pushes. Today they each rose to 17 runs (vs ~9 on a typical day), and all four sit well above the trailing daily-total average line.

---

🔗 GitHub REST API Calls Heatmap (90 days)

The calendar grid shows consumption tracks weekday PR activity, with weekends visibly lighter. The darkest cell is 2026-06-09, today's record. No multi-day sustained-high streaks are present — spikes have so far been isolated single days.

---

🍩 Top API Burners (24h)

Consumption is highly concentrated: the four reviewer workflows account for 68.8% (143K of 208K) of all REST quota, and the top 10 workflows account for 95.3%. PR Code Quality Reviewer alone is 19.2% of the day's total. This concentration means any optimisation of the reviewer agents (e.g., caching PR file lists, avoiding redundant search/GraphQL queries) would have outsized impact.

---

🔗 GitHub REST API Consumption by Workflow (last 24h)

Bars show daily cumulative core-quota per workflow; the red 15k line is the GitHub REST hourly quota for reference (not a per-run threshold). Per-run consumption is highly variable — individual PR Code Quality runs today ranged from 5 to ~4,994 core points for a near-identical ~29 requests, indicating some runs hit expensive search/GraphQL operations (which cost many quota points per request) while others did only cheap REST calls. Reducing the frequency of the expensive query paths is the clearest optimisation lever.

---

Top 10 Workflows by REST API Consumption (last 24h)

Workflow	REST API Calls	Runs	Avg Duration
PR Code Quality Reviewer	39,866	17	736.3s
Matt Pocock Skills Reviewer	38,157	17	657.3s
Test Quality Sentinel	34,226	17	556.0s
Design Decision Gate 🏗️	30,800	17	359.3s
PR Sous Chef	16,797	9	512.3s
Smoke CI	15,646	10	273.7s
PR Description Updater	6,491	8	362.1s
Auto-Triage Issues	6,172	5	357.0s
Objective Impact Report	5,690	1	412.0s
Daily Max AI Credits Test (Intentionally Fails)	4,345	3	329.0s

---

Trending Indicators

• 7-day API trend: ↑ +92.1% vs. previous 7 days
• 30-day API trend: → (insufficient data — 22 of 60 days needed for a full 30-vs-30 comparison)
• GitHub REST API call rate: ~80,183 calls/day over last 7 days (hourly limit: 15,000 — note this is a per-hour quota; daily totals legitimately exceed it as quota refreshes hourly)

---

Recommendations

1. Investigate the 17×/day reviewer cadence — four PR-gating agents each ran 17 times today. If this reflects repeated pushes to a small number of active PRs, consider debouncing or concurrency-cancelling superseded review runs.
2. Audit expensive query paths in the reviewers — per-run quota swings (5 → ~5,000) point to occasional search/GraphQL calls dominating cost. Caching PR file lists and batching lookups would cut the tail.
3. Watch for a sustained trend — today is a single-day z=3.64 spike; if 2026-06-10 stays above ~120K, treat it as a new baseline rather than an outlier.

---

📦 Cache Memory Status

• Location: /tmp/gh-aw/cache-memory/trending/api-consumption/history.jsonl
• Cache restored from previous run: yes (21 entries)
• Collection mode: incremental (history already rich; only today's 2026-06-09 entry written — the existing fuller 2026-06-08 morning snapshot was preserved, not overwritten)
• Logs start_date used: -1d (with supplemental windowed fetches to complete the 24h coverage)
• Data points stored: 22
• Earliest entry: 2026-05-18
• Retention policy: 90 days

🔍 Data quality & methodology notes

• The logs MCP logs tool repeatedly hit the 120s bridge HTTP cap; full 24h coverage required several windowed start_date/end_date fetches. Final coverage: 2026-06-08 12:11 → 2026-06-09 11:53 UTC, 445 fully-downloaded run directories.
• For today's history entry, only runs with UTC date 2026-06-09 (187 runs) were counted, keeping it comparable to prior mid-day daily snapshots.
• core_consumed is read directly from each run's run_summary.json and is independent of collection completeness, so the spike is a genuine per-run measurement, not a collection artifact.

---

Automatically generated by the api-consumption-report workflow.

Generated by 📊 GitHub API Consumption Report Agent · 453.7 AIC · ⌖ 35.5 AIC · ⊞ 12.2K · ◷

• expires on Jun 12, 2026, 4:22 AM UTC-08:00