Address residual P1+P2s from re-audit of PR #412

[igerber]

Summary

Audit follow-up to PR #412. The restored CI reviewer surfaced findings the degraded reviewer missed across all 5 prior rounds:

• P1 (REGISTRY + code comment) - the claim that "R does not ship per-path `predict_het` on placebos either, so parity is preserved by deferral" contradicts what R's `did_multiplegt_dyn(..., by_path, predict_het)` dispatcher does: it forwards `predict_het` into each per-path `did_multiplegt_main` call alongside `placebo`, so R may emit per-path placebo heterogeneity rows we do not yet mirror. Rewrite both surfaces as an explicit Python-side deferral, NOT a verified R-parity. TODO row added to track validating R's actual output and either implementing parity or documenting the deviation explicitly.

• P2 (REGISTRY rtol claim) - the per-path heterogeneity R-parity paragraph claimed `rtol ~1e-6 on point estimates AND SE`, but parity tests use `BETA_RTOL=1e-6` and `SE_RTOL=1e-5`. Split the claim and note the WLS-denominator/cohort-recentering numerical drift motivating the looser SE bound.

• P2 (replicate-weight df_survey refresh test gap) - the existing test `test_per_path_heterogeneity_replicate_weights_propagates_n_valid` would have passed if the new dedicated refresh loop failed to recompute `t_stat` / `p_value` / `conf_int` at the final `df_survey`. Strengthen to call `safe_inference(beta, se, df=df_survey)` on the first finite entry and assert the stored inference fields match.

• P2 (paths_of_interest survey gap) - the documented composability of `paths_of_interest + heterogeneity + survey_design` was not regression-locked (all survey-specific tests used `by_path=k`). Add two new tests: analytical Binder TSL coverage with selector-ordering preservation, and the multiplier-bootstrap gate under `paths_of_interest`.

No estimator behavior, weighting, variance/SE, identification, or default statistical surface changed in source - documentation accuracy plus expanded regression coverage only.

Test plan

• CI - new tests run as part of the standard slow-test matrix.
• Locally verified: `pytest tests/test_chaisemartin_dhaultfoeuille.py::TestByPathHeterogeneity::test_paths_of_interest_heterogeneity_{survey_design_analytical,survey_n_bootstrap_gate} tests/test_chaisemartin_dhaultfoeuille.py::TestByPathHeterogeneity::test_per_path_heterogeneity_replicate_weights_propagates_n_valid -m ''` all pass.

🤖 Generated with Claude Code

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 7e9f1aa861fd043ea6b276b57c7ec4363d1e5bbe

---

Overall Assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive Summary

• The diff does not change dCDH estimator logic, weighting, variance/SE computation, identification assumptions, or defaults; it changes documentation/comments and adds regression coverage around by_path heterogeneity.
• Affected methodology surface: dCDH per-path heterogeneity testing (predict_het, Web Appendix Section 1.5 / Lemma 7) and the by-path placebo-row export path.
• The registry/comment updates correctly stop claiming verified R parity for per-path placebo heterogeneity and instead document the current Python behavior as a tracked deferral. docs/methodology/REGISTRY.md:L643-L645, diff_diff/chaisemartin_dhaultfoeuille_results.py:L1862-L1903, TODO.md:L63
• The rewritten parity-tolerance language matches the actual parity test constants (BETA_RTOL=1e-6, SE_RTOL=1e-5). tests/test_chaisemartin_dhaultfoeuille_parity.py:L1404-L1469
• The new tests target the right regressions: final df_survey refresh on path_heterogeneity_effects, paths_of_interest + heterogeneity + survey_design, and the survey + multiplier-bootstrap gate under paths_of_interest. tests/test_chaisemartin_dhaultfoeuille.py:L10577-L10703
• I could not execute the tests in this environment because pytest is not installed.

Methodology

• Severity: P3. Impact: The remaining per-path placebo heterogeneity gap is now accurately documented as a Python-side deferral rather than an asserted R-parity fact; this matches the live behavior, where placebo rows in to_dataframe(level="by_path") still emit NaN het_* fields. Concrete fix: None required for approval; future parity work is already tracked. docs/methodology/REGISTRY.md:L643-L645, diff_diff/chaisemartin_dhaultfoeuille_results.py:L1862-L1903, TODO.md:L63
• No P0/P1 methodology findings.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• Severity: P3. Impact: The new TODO entry properly tracks the unresolved placebo-heterogeneity parity work, so it is mitigated under the project’s deferred-work policy rather than remaining implicit. Concrete fix: None required in this PR. TODO.md:L63

Security

• No findings.

Documentation/Tests

• No findings. The added tests are well-targeted to the touched risks, especially the dedicated final-df refresh loop in fit(). diff_diff/chaisemartin_dhaultfoeuille.py:L3957-L4070, tests/test_chaisemartin_dhaultfoeuille.py:L10577-L10703
• Verification note: test execution was not possible here because neither pytest nor python -m pytest is available in the review environment.