v0.2.0

What's new in 0.2.0

Added

• Podcast Show Summary — plain-text episode companion generated alongside the podcast transcript, with Vancouver/NLM citation, 2–3 sentence synopsis, structured "Study at a glance" bullets (adapts to study type), inline GRADE certainty ratings, and exact numbers (ORs, CIs, p-values)

Changed

• Consistent ═══ STEP N: NAME ═══ headers across all 6 pipeline steps
• Iterative correction loop redesigned: compact "Correction N of M" format, --verbose / -v flag for detailed output, best-so-far rollback, early exit after 2 consecutive quality degradations, deterministic schema_repair.py for pattern/minimum/enum violations
• Zero-value metrics display as "N/A" instead of "0.0%"

Fixed

• Correction loop headers now appear before Correcting... instead of after the previous correction's result (GH #38)
• Step 4 (Critical Appraisal) and Step 5 (Report Generation) headers now print immediately when the step starts, before any LLM call

Security

• python-dotenv bumped to >=1.2.2 — symlink following / arbitrary file overwrite
• pillow bumped to >=12.2.0 — FITS GZIP decompression bomb
• tornado pinned to >=6.5.5 — cookie injection, DoS via multipart, incomplete cookie validation
• Pygments pinned to >=2.20.0 — ReDoS via inefficient GUID regex
• requests pinned to >=2.33.0 — insecure temp file reuse

---

Full changelog