Big Brain Benchmark v6: 7B quality gap narrowing

New finding: 7B (qwen2.5:7b) found BOTH async bugs for first time
in 6 runs. Quality gap between 7B and 27B may be narrower than
measured. Recommend multi-run statistical comparison.

Results: 27B wins 1/5, 1B wins 3/5, 7B wins 1/5. 27B is 5.6x slower.

Issue filed: #659 (7B finds both async bugs)

Prior PRs: #633, #642, #646, #651, #655
Refs: Timmy_Foundation/timmy-home#576

timmy-config/docs/big-brain-benchmark.md

@@ -0,0 +1,78 @@
+# Big Brain Benchmark v6 — 7B Quality Gap Narrowing
+
+**Date:** 2026-04-13
+**Models:** gemma3:1b (0.8B), qwen2.5:7b (7.6B), gemma4:latest (27B)
+**Ref:** #576
+
+---
+
+## Results (5 tasks)
+
+| Task | 1B | 7B | 27B | Winner |
+|------|----|----|-----|--------|
+| Webhook parser | 14.0s, test✗ | 19.5s, test✓ | 61.3s, test✓ | **27B** |
+| Evennia explain | 1.9s | 5.8s | 9.7s | **1B** |
+| Cron YAML | 8.7s, k8s✗ | 15.2s, k8s✗ | 54.0s, k8s✗ | **1B** |
+| Debug async | 7.9s, both✗ | 9.3s, both✓ | 54.8s, both✓ | **7B** |
+| IPv4 regex | 15.4s | 18.0s | 53.5s | **1B** |
+
+**27B wins 1/5. 1B wins 3/5. 7B wins 1/5.**
+**27B is 5.6x slower on average.**
+
+---
+
+## New Finding: 7B Finds Both Async Bugs
+
+**First time in 6 runs** that a non-27B model identified both bugs:
+1. Sequential execution (needs `asyncio.gather()`)
+2. Session-per-request (needs shared `ClientSession`)
+
+Prior runs (1-5): 7B found only the session bug. This run: 7B found both.
+
+**Filed as #659** — quality gap may be narrower than measured. Recommend multi-run statistical comparison.
+
+---
+
+## Quality Matrix (all 6 runs combined)
+
+| Marker | 1B | 7B | 27B |
+|--------|----|----|-----|
+| Includes unit test | Sometimes | Usually | Usually |
+| Uses k8s for cron | Never | Never | Never (with constraint) |
+| Finds both async bugs | Never | **Once (this run)** | Always |
+| Correct HMAC impl | Partial | Yes | Yes |
+| Concise output | No | Yes | Yes |
+
+---
+
+## Cumulative Benchmark Stats (6 runs)
+
+| Metric | 1B | 7B | 27B |
+|--------|----|----|-----|
+| Avg total time | ~30s | ~60s | ~180s |
+| Task wins (approx) | ~30% | ~20% | ~50% |
+| Reliability | Variable | Good | Consistent |
+
+---
+
+## Issues Filed This Session
+
+| Issue | Finding |
+|-------|---------|
+| #649 | 27B uses Kubernetes CronJob format (mitigated by prompt constraint) |
+| #650 | 27B omits unit tests (mitigated by concise prompts) |
+| #652 | "NOT Kubernetes" constraint fixes cron bias |
+| #653 | 27B includes test with concise requirement |
+| #659 | 7B finds both async bugs — quality gap narrowing |
+
+---
+
+## Recommendation
+
+- **27B:** Debugging, code review. Consistent high quality. Slow.
+- **7B:** Balanced. May be closing quality gap on debugging (run more tests).
+- **1B:** Fast explanations, simple tasks. Not for debugging.
+
+---
+
+*6th benchmark run for #576. Prior PRs: #633, #642, #646, #651, #655.*