Overall Assessment

✅ Looks good. No unmitigated P0/P1 findings.

Executive Summary

• Docs-only PR; no estimator code, defaults, variance paths, or inference helpers changed.
• The review’s core CBWSDID summary matches the arXiv source: two-stage design adjustment plus Wing-style corrective aggregation, repeated-episode extension, and R package reference. (arxiv.org)
• Assumptions, edge cases, and inference caveats are substantially captured, including overlap/nondegeneracy, pre-treatment-only design weights, unit-level clustering, conditional-on-weights SEs, and nonsmooth matching bootstrap caveat. (arxiv.org)
• Existing StackedDiD registry/code context was checked; the review appropriately flags the count-convention and regression-structure decisions as PR-B work.
• One non-blocking documentation clarity item is noted below.

Methodology

Finding: Mixed κ_pre notation remains in implementation-facing prose
Severity: P3
Location: docs/methodology/papers/ustyuzhanin-2026-review.md:L48, docs/methodology/papers/ustyuzhanin-2026-review.md:L126, clarified later at docs/methodology/papers/ustyuzhanin-2026-review.md:L89 and L257
Impact: The review correctly documents that the paper itself mixes sign conventions, but the early “Key implementation requirements” still use {−κ_pre,…,κ_post} while later adopting signed negative κ_pre. A future implementation could copy the early formula without reading the later caveat.
Concrete fix: No blocker for this PR. For clarity, normalize implementation-facing lines to the chosen signed convention, e.g. {κ_pre,…,κ_post}, and keep the source-paper inconsistency only in the “Gaps and Uncertainties” discussion.

Code Quality

No findings. No executable code changed.

Performance

No findings. No performance-sensitive code changed.

Maintainability

No blocking findings. The PR clearly marks REGISTRY, references, R parity, and implementation wiring as PR-B scope in docs/methodology/papers/ustyuzhanin-2026-review.md:L12.

Tech Debt

No blocking findings. Deferred implementation/testing work is explicitly scoped in the review’s PR-B checklist at docs/methodology/papers/ustyuzhanin-2026-review.md:L202-L211.

Security

No findings. The added file is methodology prose and does not introduce secrets, credentials, or executable workflow changes.

Documentation/Tests

No blocking findings. Test additions are not required for this docs-only paper review. I also checked the doc-deps integrity test; it enforces stale referenced paths and unmapped public diff_diff/**/*.py modules, not every standalone methodology markdown file, so omitting docs/doc-deps.yaml is not a CI issue for this PR.

🔁 AI review rerun (requested by @igerber)

Head SHA: c89f9fd1d4b5bba0ec118e98a70d3098cbb53bec

Overall Assessment

⚠️ Needs changes. One unmitigated P1: the PR removes the SyntheticControl CWZ conformal inference public surface without a compatibility path or clear removal/deferment contract.

Executive Summary

• CBWSDID review is docs-only and broadly matches the arXiv source at the abstract level: design weights within sub-experiments, Wing-style aggregation, repeated 0→1/1→0 extension, and an R package reference. (arxiv.org)
• The PR is not just a CBWSDID paper-review change; it deletes diff_diff/conformal.py, conformal methods, diagnostics, docs, and tests.
• P1: current callers of SyntheticControlResults.conformal_test() / CI helpers now fail with missing attributes instead of a documented migration/deprecation.
• P2/P3 docs drift remains around DiagnosticReport reporting and the CBWSDID κ_pre convention.
• No security or performance issues found.

Methodology

Finding: Removed CWZ conformal inference surface without an explicit methodology/removal contract
Severity: P1
Location: diff_diff/conformal.py deleted; diff_diff/synthetic_control_results.py:L1810-L1823; docs/methodology/papers/chernozhukov-wuthrich-zhu-2021-review.md:L100-L106
Impact: This removes an implemented SyntheticControl inference method family rather than changing its math. Existing users get AttributeError for conformal_test, conformal_confidence_intervals, conformal_average_effect, and accessors, with no deprecation stub or clear “feature deferred/removed” registry note.
Concrete fix: Either restore the conformal module/methods/tests/docs, or add an explicit removal/deferment path: changelog entry, registry/paper-review status update, and compatibility methods that raise a clear NotImplementedError with the supported alternative or planned replacement.

Finding: Mixed κ_pre convention remains in implementation-facing CBWSDID prose
Severity: P3
Location: docs/methodology/papers/ustyuzhanin-2026-review.md:L48, L126, clarified later at L89 and L258
Impact: The review correctly flags the paper’s notation inconsistency, but early checklist-style prose still uses {−κ_pre,…} while later recommending signed negative κ_pre. A future implementation could copy the wrong convention.
Concrete fix: Normalize implementation-facing formulas/checklists to {κ_pre,…,κ_post} with κ_pre < 0, and keep {−κ_pre,…} only inside the explicit source-ambiguity discussion.

Code Quality

Finding: Public API removal produces raw missing-method failures
Severity: P1
Location: diff_diff/synthetic_control_results.py:L1810-L1823; deleted diff_diff/conformal.py
Impact: Consumers using the prior results API fail at attribute lookup, not with a controlled, documented error.
Concrete fix: Restore the methods or leave tombstone methods for one release that raise a clear NotImplementedError.

Performance

No findings. The removed conformal code reduces runtime surface; no new performance-sensitive code was added.

Maintainability

Finding: Reporting methodology doc no longer matches DiagnosticReport behavior
Severity: P2
Location: docs/methodology/REPORTING.md:L266-L274; implementation still emits confidence_set at diff_diff/diagnostic_report.py:L2462-L2504
Impact: The edited reporting note omits the still-supported Firpo-Possebom confidence_set block, so future maintainers may think it is not part of native SCM diagnostics.
Concrete fix: Re-add confidence_set to the SyntheticControlResults native-diagnostics list while keeping conformal omitted if that rollback is intentional.

Tech Debt

Finding: Conformal follow-up TODO was removed together with the implementation
Severity: P3
Location: TODO.md:L88-L89; docs/methodology/papers/chernozhukov-wuthrich-zhu-2021-review.md:L129-L143
Impact: If CWZ conformal inference is now deferred rather than abandoned, the remaining work is no longer tracked.
Concrete fix: Add a replacement TODO saying CWZ conformal inference is deferred, or update the CWZ paper review to state it is no longer planned.

Security

No findings. No secrets, credentials, or new executable workflow risk observed.

Documentation/Tests

Finding: CBWSDID review claims this PR touches only the review file
Severity: P3
Location: docs/methodology/papers/ustyuzhanin-2026-review.md:L12
Impact: The statement is false for this diff, which changes code, tests, docs, registry, references, and TODO.
Concrete fix: Rephrase to “this review artifact is documentation-only; implementation is deferred,” removing the PR-scope claim.

Finding: Conformal tests were deleted with the feature
Severity: P3 if rollback is intentional; P1 if conformal is meant to remain
Location: tests/test_methodology_synthetic_control.py:L3521-L3540; tests/test_diagnostic_report.py:L2114-L2135
Impact: If conformal is restored, the methodology oracle and diagnostic tests must come back with it.
Concrete fix: Restore the deleted tests alongside the implementation, or document the intentional removal/deferment.

Path to Approval

1. Resolve the conformal removal: restore the conformal implementation/tests/docs, or add explicit removal/deferment documentation plus compatibility tombstone methods with clear errors.
2. Update docs/methodology/REPORTING.md so it still lists the existing confidence_set native diagnostic.
3. Normalize the CBWSDID κ_pre implementation-facing notation to the signed convention.

🔁 AI review rerun (requested by @igerber)

Head SHA: 0b396acc13e3df05e4eeba625ee7dbf3fc4777b3

Overall Assessment

✅ Looks good. No unmitigated P0/P1 findings in the current diff.

Executive Summary

• Re-review scope is now docs-only: the current diff only adds docs/methodology/papers/ustyuzhanin-2026-review.md.
• The prior conformal-inference P1 is no longer present in this diff scope; existing registry/TODO conformal entries remain intact at docs/methodology/REGISTRY.md:L2006-L2041 and TODO.md:L89.
• The CBWSDID review is broadly faithful to the arXiv source: design weights within sub-experiments, Wing-style corrective aggregation, repeated 0→1 / 1→0 episodes, unit clustering, and bootstrap caveats are represented consistently. (arxiv.org)
• One minor documentation consistency issue remains around κ_pre notation, but it is explicitly framed as source ambiguity and does not block approval.

Methodology

Finding: Mixed κ_pre notation remains in implementation-facing formulas
Severity: P3
Location: docs/methodology/papers/ustyuzhanin-2026-review.md:L48, docs/methodology/papers/ustyuzhanin-2026-review.md:L126; clarified at L89 and L258
Impact: The review correctly flags the paper’s internal ambiguity: the paper describes {−κ_pre,…,κ_post} in some places while examples use signed negative windows like κ=(-3,2), and the regression sum starts at h=κ_pre. (arxiv.org) A future implementer could still copy the early {−κ_pre,…} formulas despite the later signed-convention guidance.
Concrete fix: Normalize implementation-facing formulas to e ∈ {κ_pre,…,κ_post} where κ_pre < 0, and keep {−κ_pre,…} only inside the explicit source-ambiguity discussion.

Code Quality

No findings. This PR adds documentation only and does not change estimator code, inference code, defaults, or public APIs.

Performance

No findings. No executable or performance-sensitive code is introduced.

Maintainability

No blocking findings. The review explicitly marks packaging, StackedDiD integration, FE-vs-means equivalence, R parity, and unbalanced-panel count conventions as PR-B decisions, which is appropriate for a paper-review artifact.

Tech Debt

No blocking findings. The deferred implementation work is documented in the new review itself as PR-B scope at docs/methodology/papers/ustyuzhanin-2026-review.md:L202-L211 and L245-L250.

Security

No findings. No secrets, credentials, executable workflows, or data files are added.

Documentation/Tests

Finding: Validation not runnable in this review environment
Severity: P3
Location: tests/test_doc_deps_integrity.py
Impact: I attempted the lightweight doc-deps integrity test, but pytest is not installed in this environment (python -m pytest reports No module named pytest). This leaves CI validation to the PR pipeline.
Concrete fix: No PR change required if CI runs normally; otherwise run tests/test_doc_deps_integrity.py in the project test environment.