Add EfficientDiD v2.7 LinkedIn carousel

[igerber]

Summary

• Add 8-slide dark-themed LinkedIn carousel PDF generator for v2.7 release
• Highlights EfficientDiD as the first open-source implementation of Chen, Sant'Anna & Xie (2025)
• Slides: hook, problem, insight, PT-All assumption explainer, efficiency payoff, API code, safety net, CTA
• Follows established carousel pattern from v2.3–v2.6 (FPDF subclass, matplotlib equation/diagram rendering)

Methodology references (required if estimator / math changes)

• N/A — no methodology changes, marketing asset only

Validation

• Tests added/updated: No test changes (marketing asset)
• Backtest / simulation / notebook evidence: N/A
• Manual verification: 8 pages, 270×337.5mm (4:5 aspect), equations render correctly, no footer overlap

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

⚠️ Needs changes — no estimator code changed, but the new carousel overstates the documented PT-Post equivalence between EfficientDiD and CallawaySantAnna.

Executive Summary

• PR scope is marketing-only: a new generator script plus a committed PDF artifact.
• I verified the committed PDF metadata is consistent with the PR description: 8 pages and a 270×337.5 mm media box.
• P1 methodology issue: slides 4 and 6 broaden PT-Post equivalence to an unqualified “Callaway-Sant'Anna” / “identical” claim, while the repo’s own methodology/docs/tests scope the exact match to post-treatment ATT(g,t) and note pre-treatment diagnostics can differ.
• P3 documentation issue: slide 1’s “First & Only Open-Source Implementation” claim is not supported anywhere in the in-repo methodology/docs.
• Minor P3 maintainability issue: internal slide numbering drifts from the actual 8-page deck starting mid-file.

Methodology

• Severity: P1. Impact: The carousel turns a documented post-treatment equivalence into a blanket equivalence claim. Slide 4 equates PT-Post with “= Callaway-Sant'Anna” at carousel/generate_v27_pdf.py:459, and slide 6 says “Under PT-Post, reduces exactly to Callaway-Sant'Anna” plus identical ATT/SE/CI rows at carousel/generate_v27_pdf.py:626 and carousel/generate_v27_pdf.py:662. The project’s own source material is narrower: the registry documents PT-Post as a just-identified single-baseline estimator at docs/methodology/REGISTRY.md:493 and docs/methodology/REGISTRY.md:605, the API docs say PT-Post “matches CS for post-treatment effects” at docs/api/efficient_did.rst:12 and docs/api/efficient_did.rst:97, the tutorial explicitly says pre-treatment diagnostics may differ at docs/tutorials/15_efficient_did.ipynb:164, and tests only assert exact equality for post-treatment ATT(g,t) at tests/test_efficient_did.py:246 and tests/test_efficient_did.py:909. Concrete fix: rewrite the PT-Post copy to say “matches Callaway-Sant'Anna for post-treatment ATT(g,t)” (ideally naming control_group='never_treated'), and either remove the generic identical SE/CI table or back it with an in-repo citation/test before regenerating the PDF.

Code Quality

• No findings. The script parses successfully. I could not fully execute it in this sandbox because matplotlib is not installed here.

Performance

• No findings. This does not touch library hot paths or estimator runtime.

Maintainability

• Severity: P3. Impact: Internal slide numbering is off by one from slide 5 onward, starting at carousel/generate_v27_pdf.py:534 and continuing through carousel/generate_v27_pdf.py:709. That is minor, but it repeats a pattern already called out in CHANGELOG.md:61. Concrete fix: rename the methods/docstrings to the actual slide numbers, or switch to title-based names only.

Tech Debt

• No findings. Existing EfficientDiD TODO items at TODO.md:48 and TODO.md:49 are unrelated and do not mitigate the methodology wording issue above.

Security

• No findings. I did not see secrets or PII in the new script or the committed PDF metadata.

Documentation/Tests

• Severity: P3. Impact: Slide 1’s “First & Only Open-Source Implementation” claim at carousel/generate_v27_pdf.py:333 is an external exclusivity claim with no supporting note in the EfficientDiD registry or API docs, so it is brittle and hard to verify in review. Concrete fix: either add a specific citation/supporting note for exclusivity, or soften it to a source-backed statement such as “Open-source Python implementation of EfficientDiD.”
• Validation note: the committed PDF advertises /Count 8 and /MediaBox [0 0 765.35 956.69], which matches the PR’s stated 8 pages and 270×337.5 mm size.

Path to Approval

1. Narrow the PT-Post messaging on slides 4 and 6 to post-treatment ATT(g,t) equivalence only, rather than “Callaway-Sant'Anna” generically.
2. Remove or explicitly substantiate the “identical Standard error / 95% CI” claims if you want to keep that table.
3. Regenerate carousel/diff-diff-v27-carousel.pdf from the updated script.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 784f067478bdd2f87bfc0c1b75edb447f99f253e

---

Overall Assessment

✅ Looks good — this re-review resolves the prior methodology blocker, and I found no unmitigated P0/P1 findings in the PR. One P3 informational issue remains on slide 1’s exclusivity claim.

Executive Summary

• Re-review result: the prior PT-Post overstatement is fixed. Slides 4 and 7 now limit the exact-equivalence claim to post-treatment ATT(g,t) in carousel/generate_v27_pdf.py:L455-L470 and carousel/generate_v27_pdf.py:L615-L648, which matches docs/methodology/REGISTRY.md:L493-L494, docs/api/efficient_did.rst:L12-L13, and tests/test_efficient_did.py:L909-L924. citeturn3view0
• The earlier slide-number drift is fixed; the generator now defines slides 1 through 8 in order at carousel/generate_v27_pdf.py:L324-L696.
• No estimator, weighting, variance/SE, identification-assumption logic, or default behavior changed. This PR is limited to a carousel generator plus its checked-in PDF.
• P3 only: slide 1 still says “First Open-Source Implementation” in carousel/generate_v27_pdf.py:L324-L362. For diff-diff 2.7.0 dated 2026-03-15 in CHANGELOG.md:L10, that is not a safe claim: a public MIT-licensed edid implementation of the same paper was already public on 2026-01-25. citeturn18open0turn10open0turn13open0
• Validation note: the source parses cleanly via compile(), and the committed PDF metadata shows 8 pages with MediaBox [0 0 765.35 956.69], consistent with the script’s 270×337.5 mm canvas in carousel/generate_v27_pdf.py:L20-L22.

Methodology

• Severity: none. Impact: the prior blocker is resolved. The PT-Post language is now scoped to post-treatment ATT(g,t) only, which is consistent with Corollary 3.2 and the repo’s own registry/API/tests. Concrete fix: none. carousel/generate_v27_pdf.py:L455-L470 carousel/generate_v27_pdf.py:L615-L648 docs/methodology/REGISTRY.md:L493-L494 docs/api/efficient_did.rst:L97-L148 tests/test_efficient_did.py:L246-L261 tests/test_efficient_did.py:L909-L924. citeturn3view0
• Severity: none. Impact: slide 5’s “often exceeding 40% gains in precision” language is source-backed rather than a methodology deviation. Concrete fix: none. carousel/generate_v27_pdf.py:L533-L571. citeturn19view0

Code Quality

• Severity: none. Impact: the Python source parses successfully, and I did not find new anti-patterns or correctness issues in the reviewed script. Concrete fix: none.

Performance

• Severity: none. Impact: this PR does not touch estimator hot paths; the work is confined to offline PDF rendering. Concrete fix: none.

Maintainability

• Severity: none. Impact: the earlier internal slide-number naming drift is resolved; the method/docstring sequence is now consistent from slide_01_hook() through slide_08_cta(). Concrete fix: none. carousel/generate_v27_pdf.py:L324-L696

Tech Debt

• Severity: none. Impact: no new deferred library limitation was introduced that needs TODO.md tracking. Concrete fix: none.

Security

• Severity: none. Impact: I did not find secrets, PII, or risky untrusted execution paths in the added script/PDF. Concrete fix: none.

Documentation/Tests

• Severity: P3. Impact: slide 1’s “First Open-Source Implementation” claim is not defensible for this release and is easily challengeable by readers. Concrete fix: replace it with a source-backed statement such as “diff-diff v2.7 adds EfficientDiD” or “Open-source EfficientDiD implementation in diff-diff,” then regenerate carousel/diff-diff-v27-carousel.pdf. carousel/generate_v27_pdf.py:L324-L362 CHANGELOG.md:L10. citeturn18open0turn10open0turn13open0
• Severity: none. Impact: no estimator-test updates were required because the PR is marketing-only, and the checked-in PDF metadata is internally consistent with the generator. Concrete fix: none. carousel/generate_v27_pdf.py:L20-L22