[cache-strategy] Daily Cache Strategy Analysis - 2026-09-02

[github-actions[bot]]

📊 Daily cache strategy analysis - 2026-09-02

Summary

Analyzed 74 workflow runs from the last 24 hours that declared cache-memory: true.

• 74/74 runs showed cache-miss indicators or did not reach a successful Update cache-memory step
• 40+ workflows matched a volatile cache-key pattern without restore-key fallbacks at the prompt level
• 2 new issues were opened for the most actionable untracked offenders
• Several previously known offenders continued to regress today

New issues created

• [cache-strategy] Fix cache miss in AI Moderator #58004 — [cache-strategy] Fix cache miss in AI Moderator
• [cache-strategy] Fix cache miss in Code Scanning Fixer #58005 — [cache-strategy] Fix cache miss in Code Scanning Fixer

Existing tracked issues hit again today

• #aw_go_test — Daily Go Test Parallelizer
• #aw_deeprep — Deep Report
• #aw_sessins — Copilot Session Insights
• #aw_issuecache — Daily Issues Report Generator
• #aw_compqc — Daily Compiler Quality Check

Highest-priority findings

Critical

• AI Moderator — 10 runs today, all misses or missing cache updates; volatile-key pattern detected
• Code Scanning Fixer — 2 runs today, both misses; volatile-key pattern detected
• Daily Go Test Parallelizer — 8 runs today, all misses; already tracked in #aw_go_test
• Deep Report — 2 runs today, both misses; already tracked in #aw_deeprep

Medium / trend-based

• Daily Compiler Quality Check — miss streak increased to 3 days, 14-day miss rate reached 50%; already tracked in #aw_compqc
• Copilot Session Insights — 14-day miss rate remained at 50%; already tracked in #aw_sessins
• Daily Issues Report Generator — 14-day miss rate remained at 50%; already tracked in #aw_issuecache

Evidence snippets

Representative cache signals seen in analyzed runs

• prompt references to missing_data with reason: "cache_memory_miss"
• prompt references to /tmp/gh-aw/cache-memory/
• prompt patterns indicating run-specific keys such as run_id, run_number, timestamp, or workflow-run-id without restore fallback wording
• run_summary.json agent jobs where Update cache-memory was absent or not successful

Representative runs:

• AI Moderator: 33625933588, 33626194781, 33626744963, 33626889433, 33627547476, 33634261395, 33636753063, 33641758968, 33661969597, 33663214863
• Daily Go Test Parallelizer: 33584892977, 33591456221, 33600572141, 33609751734, 33619943619, 33631387025, 33642641896, 33655656497
• Deep Report: 33599982994, 33630641804
• Code Scanning Fixer: 33601555914, 33632778572
• Daily Compiler Quality Check: 33584903739

Recommendations

1. Standardize the valid "last one wins" pattern: allow per-run save keys only when paired with restore-key fallbacks.
2. Require explicit cache hit/miss logging in cache-enabled workflows.
3. Ensure cache-enabled workflows always read existing state before recomputing.
4. Ensure the cache update step runs on successful or partially successful analysis paths.
5. Consider linting workflow prompts for volatile cache-key patterns without restore fallbacks.

Generated by 📊 Daily Cache Strategy Analyzer · pi · gpt54 · 42.1 AIC · ⌖ 9.27 AIC · ⊞ 8K · ◷

• expires on Sep 3, 2026, 10:50 AM UTC-08:00