Add v2.6 LinkedIn carousel PDF

[igerber]

Summary

• Add carousel PDF generator script (carousel/generate_v26_pdf.py) for the v2.6 continuous DiD release
• Generate 10-slide LinkedIn carousel covering: hook, problem (binary DiD bias), solution (3-step procedure), math (B-spline OLS / ATT / ACRT), dose-response curve, ATT vs ACRT comparison, assumptions (PT vs SPT), code example, use cases, full toolkit grid, and CTA
• Includes matplotlib equation rendering, dose-response figure, and light gradient backgrounds

Methodology references (required if estimator / math changes)

• N/A - no methodology changes (marketing carousel only)

Validation

• Tests added/updated: No test changes (standalone PDF generator)
• Backtest / simulation / notebook evidence (if applicable): N/A

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment: ⚠️ Needs changes

Executive Summary

• Methodology mismatch in the assumptions slide: Standard PT is stated as identifying ATT(d), but the registry says PT only identifies ATT(d|d)/ATT^{loc}; SPT is required for ATT(d)/ACRT(d).
• Estimation step text reverses the regression direction (“dose on outcomes”), conflicting with the registry’s OLS specification.
• No estimator or inference code paths changed; edge‑case checklist items are not triggered by this PR.
• Minor maintainability: slide numbering comments/docstrings are inconsistent.

Methodology

• P1 — ContinuousDiD identification is misstated under Standard PT. Impact: misleading claims about what is identified; could cause users to over‑interpret ATT(d) under PT. Fix: change the Standard PT row to reference ATT(d|d) (and optionally ATT^{loc}), and keep ATT(d)/ACRT(d) under Strong PT. References: carousel/generate_v26_pdf.py:L729-L742, docs/methodology/REGISTRY.md:L408-L430.
• P1 — Estimation step reverses the regression direction. Impact: incorrect methodological description (implies regressing dose on outcomes rather than outcomes on dose basis). Fix: update to “Fit B‑spline regression of demeaned outcomes on dose (basis).” References: carousel/generate_v26_pdf.py:L406-L409, docs/methodology/REGISTRY.md:L432-L436.

Code Quality

• No issues found.

Performance

• No issues found.

Maintainability

• P3 — Slide numbering comments/docstrings are inconsistent (e.g., Slide 8 label vs “Slide 7” docstring; Slide 8/9 mismatch). Impact: confusion when editing/reordering slides. Fix: align comments/docstrings with actual slide order. References: carousel/generate_v26_pdf.py:L808-L869.

Tech Debt

• No issues found.

Security

• No issues found.

Documentation/Tests

• No issues found.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 6626ba01c21ae22eae648cd3eee6f43fab3befa6

---

Overall Assessment ✅ Looks good

Executive Summary

• Methodology content for ContinuousDiD in the carousel (ATT/ACRT equations, PT/SPT identification, IF+bootstrap) matches the Methodology Registry; no estimator/inference code changes.
• Generator script is self-contained and outputs the new 11-page PDF in carousel/.
• P3: temp image cleanup isn’t guaranteed on exceptions.
• P3: dependencies to run the script aren’t documented or declared.
• Tests not run (not requested; change is a standalone asset/script).

Methodology

• No findings. Affected method: ContinuousDiD. Carousel equations/assumptions are consistent with the registry and in-code docs. carousel/generate_v26_pdf.py:L488-L806

Code Quality

• Severity: P3 | Impact: Temporary PNG files may persist if rendering fails before cleanup() runs, leaving clutter in temp directories. | Fix: wrap slide rendering/output in try/finally (or register atexit) to always call cleanup(). | Location: carousel/generate_v26_pdf.py:L49-L92, carousel/generate_v26_pdf.py:L1049-L1068

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• No findings.

Security

• No findings.

Documentation/Tests

• Severity: P3 | Impact: Running the generator requires matplotlib, Pillow, and fpdf, but there’s no install/run guidance; contributors may be unable to regenerate the PDF. | Fix: add a short carousel/README.md (or a carousel optional dependency group in pyproject.toml) with install/run steps. | Location: carousel/generate_v26_pdf.py:L9-L15