Quality overhaul: bug fixes, tests/CI, algorithms, newsletter UX, docs

[bits-bytes-nn]

Summary

End-to-end quality overhaul across code, tests, algorithms, newsletter UX, and docs.

Bug fixes

• Content-sufficiency gate — drop thin posts before summarization (fixes "article too short").
• Crawler health tracking (OK/EMPTY/FAILED + CrawlReport) with SNS alerts; updated stale source URLs; removed abandoned feeds (ncsoft, huyenchip).
• Lambda-no-Chrome fail-fast guard; Dockerfiles bumped to Python 3.12.

Code / architecture / AWS

• Removed dead async batch path; resilient HTTP layer (header rotation, 429 backoff w/ Retry-After).
• Least-privilege IAM (scoped S3/SES/SNS/SSM/Bedrock/Logs; cross-region-safe; fail-closed bucket).
• Prompt caching (backend-aware cachePoint vs cache_control); configurable thinking budgets w/ clamp.
• max_posts capped before summarization (cost); fixed sequential-fallback result misalignment.

Algorithms

• Filtering rubric tuned to reproducible anchor scores (filtering thinking OFF → deterministic).
  Live-verified: determinism 100%, on-grid 100%, band-match 100%.

Tests / CI

• ~205 offline pytest tests; GitHub Actions CI (ruff, ruff-format, mypy, pytest, eval dry-run, cdk synth).

Docs

• assets/tech-doc.md (line-by-line, Korean) + Excalidraw pipeline + draw.io AWS architecture diagrams.

🤖 Generated with Claude Code