Overall Assessment: ⚠️ Needs changes

Executive Summary

• Doc-only PR adds the TwoStageDiD tutorial notebook and updates the tutorial list.
• Two methodology statements in the new notebook conflict with the Methodology Registry/implementation and should be corrected (CallawaySantAnna inference and treatment_effects weights).
• No code paths changed; edge-case checklist is not applicable beyond documentation accuracy.
• Tests not run (doc-only).

Methodology

• Severity: P1. Impact: The notebook states CallawaySantAnna “standard errors come from an analytical multiplier bootstrap,” which misrepresents the default inference path (analytical influence-function SEs by default; multiplier bootstrap is optional). This can mislead users comparing inference across estimators. Fix: Replace with “Analytical influence-function SEs by default; optional multiplier bootstrap when n_bootstrap > 0.” Location: docs/tutorials/12_two_stage_did.ipynb:L131-L133.
• Severity: P1. Impact: The notebook claims treatment_effects.weight is 1/n_treated, but the implementation assigns 1/n_valid for finite tau_hat and 0 for NaN rows, which matters in rank-deficient/NaN cases. Users doing custom aggregation could apply incorrect weights. Fix: Update text to “weight = 1/n_valid for finite tau_hat (0 for NaN)” or explicitly note NaN handling. Location: docs/tutorials/12_two_stage_did.ipynb:L100-L106 (implementation: diff_diff/two_stage.py:L778-L785).

Code Quality
No findings.

Performance
No findings.

Maintainability
No findings.

Tech Debt
No findings.

Security
No findings.

Documentation/Tests

• Severity: P3. Impact: The new notebook’s execution is not validated in CI, so future regressions could slip in unnoticed. Fix: Consider adding notebook execution to CI or document a local validation step (e.g., jupyter execute) in contributor docs. Location: docs/tutorials/12_two_stage_did.ipynb.

/ai-review

🔁 AI review rerun (requested by @igerber)

Head SHA: 9cdfa17a139ea12245acd52cd14b4e10ea15cd79

Overall assessment: ⚠️ Needs changes

Executive summary

• Doc-only PR; no estimator or inference code changes.
• Tutorial contains an accuracy issue: it says per-observation treatment_effects is unique to TwoStageDiD and that ImputationDiD lacks it, which contradicts the existing ImputationDiD API.
• Update the tutorial text and summary table to reflect that ImputationDiD also exposes treatment_effects.

Methodology

• No findings.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• No findings.

Security

• No findings.

Documentation/Tests

• Severity: P2. Impact: The tutorial claims treatment_effects is unique to TwoStageDiD and the summary table says ImputationDiD has no per-observation effects, but ImputationDiDResults includes treatment_effects. This misleads users about available outputs. Fix: Update the “Per-Observation Treatment Effects” section to remove “unique” and adjust the summary table to show ImputationDiD = Yes (or explicitly describe differences in weights/variance). Location: docs/tutorials/12_two_stage_did.ipynb:L97-L119, docs/tutorials/12_two_stage_did.ipynb:L238-L240. Evidence: diff_diff/imputation.py:L90-L98.
• Severity: P3. Impact: Tests not run (docs-only). Fix: If you want execution assurance, run the notebook via the standard doc execution flow (e.g., jupyter execute). Location: N/A.

/ai-review

🔁 AI review rerun (requested by @igerber)

Head SHA: 113354e73eaef89a386a6d360a02955c9141bd5f

Overall Assessment
✅ Looks good

Executive Summary

• Adds a new tutorial notebook at docs/tutorials/12_two_stage_did.ipynb and updates the tutorial index in CLAUDE.md.
• Methodology cross‑check: TwoStageDiD two‑stage procedure, point‑estimate identity with ImputationDiD, and GMM sandwich variance statements align with docs/methodology/REGISTRY.md and diff_diff/two_stage.py.
• No estimator or inference code changes; edge‑case checklist not triggered.
• Tests not run (doc‑only change).

Methodology

• No findings.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• No findings.

Security

• No findings.

Documentation/Tests

• No findings.