Bump version to 2.7.4

[igerber]

Summary

• Bump version to 2.7.4 across __init__.py, pyproject.toml, rust/Cargo.toml, and docs/llms-full.txt
• Add CHANGELOG entry for v2.7.4: survey/sampling weights support, EfficientDiD validation tests, simulation-based power analysis, and numerous survey-related fixes

Methodology references (required if estimator / math changes)

• N/A - version bump only, no methodology changes in this PR

Validation

• No test changes (version bump only)

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good

Executive Summary

• The diff is limited to release metadata and documentation: CHANGELOG.md:L10-L36, diff_diff/init.py:L190-L190, pyproject.toml:L7-L7, rust/Cargo.toml:L3-L3, and docs/llms-full.txt:L5-L5.
• I found no estimator, weighting, variance/SE, identification, or default-behavior code changes in this PR.
• The methodology-sensitive changelog items are generally backed by code and registry material already present in-tree, including survey support and power-analysis notes in diff_diff/survey.py:L1-L15, docs/methodology/REGISTRY.md:L1712-L1723, and docs/methodology/REGISTRY.md:L1888-L1911.
• One documentation overstatement remains: the new changelog says simulation-based power analysis supports “all estimators,” but the implementation still relies on a finite registry and requires custom hooks for unregistered estimators; see CHANGELOG.md:L22-L25, diff_diff/power.py:L446-L535, and diff_diff/power.py:L1522-L1529.
• Version strings are otherwise synchronized across the touched artifacts.

Methodology

• P3 Informational. Impact: No methodology-affecting code changed in this PR. The release-note items about survey inference, power-analysis edge cases, and FWL-related absorb behavior align with existing documented implementation in docs/methodology/REGISTRY.md:L1712-L1723, docs/methodology/REGISTRY.md:L1888-L1911, and diff_diff/survey.py:L1-L15. Concrete fix: None.

Code Quality

• P3 Informational. Impact: The version bump is internally consistent across diff_diff/init.py:L190-L190, pyproject.toml:L7-L7, rust/Cargo.toml:L3-L3, and docs/llms-full.txt:L5-L5. Concrete fix: None.

Performance

• P3 Informational. Impact: No executable-path or hot-path code changed in this diff. Concrete fix: None.

Maintainability

• P3 Informational. Impact: Change scope is narrow and coherent; no new maintainability risk was introduced beyond the existing manual version synchronization pattern. Concrete fix: None.

Tech Debt

• P3 Informational. Impact: This PR does not introduce a new deferred limitation that needs tracking in TODO.md. Concrete fix: None.

Security

• P3 Informational. Impact: No secrets, credentials, or attack-surface changes are present in the touched files. Concrete fix: None.

Documentation/Tests

• P2 Documentation accuracy. Impact: The changelog overstates power-analysis coverage by saying “Simulation-based power analysis for all estimators” and “support all estimators with result_extractor” in CHANGELOG.md:L22-L25. The actual implementation only has built-in support for the registry-listed estimators in diff_diff/power.py:L446-L535, and for unregistered estimators it still raises unless the caller supplies a custom data_generator plus fit kwargs in diff_diff/power.py:L1522-L1529. That can mislead users into expecting out-of-the-box support for unsupported estimators. Concrete fix: Narrow the changelog wording to match the contract, e.g. “support for all registry-backed estimators” or “support for unregistered estimators when a custom data_generator, estimator_kwargs, and optionally result_extractor are provided.”
• P3 Informational. Impact: The EfficientDiD/HRS validation claims in the new changelog are backed by existing tests and provenance docs in tests/test_efficient_did_validation.py:L1-L24 and tests/data/README.md:L3-L33. Concrete fix: None.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 0e42937043635976106ad4097b43aad7a0cc688a

---

Overall Assessment

✅ Looks good

Executive Summary

• Re-review scope remains limited to release metadata and documentation changes in CHANGELOG.md:L10-L36, diff_diff/init.py:L190-L190, pyproject.toml:L7-L7, rust/Cargo.toml:L3-L3, and docs/llms-full.txt:L5-L5.
• No estimator implementation, weighting logic, variance/SE computation, identification checks, or default behaviors changed in this diff.
• The prior documentation concern about simulation-power scope is resolved: the changelog now limits built-in support to “registry-backed estimators,” matching the finite registry in diff_diff/power.py:L446-L535 and the public simulate_power() contract in diff_diff/power.py:L1419-L1463.
• The methodology-sensitive release notes are backed by existing registry/code/test material for survey inference, FWL guardrails, and EfficientDiD validation in docs/methodology/REGISTRY.md:L1849-L1911, diff_diff/estimators.py:L157-L188, diff_diff/estimators.py:L263-L270, diff_diff/twfe.py:L62-L85, tests/test_efficient_did_validation.py:L1-L230, and tests/data/README.md:L3-L54.
• Version strings and the new compare link are synchronized across the touched artifacts.

Methodology

• Severity: P3. Impact: No methodology-affecting code changed in this PR. The new release-note claims about survey_design, Taylor-series linearization, weight types, FPC, and lonely-PSU handling match the existing estimator signatures, survey module references, and registry notes in CHANGELOG.md:L13-L18, diff_diff/estimators.py:L157-L188, diff_diff/twfe.py:L62-L85, diff_diff/survey.py:L1-L15, and docs/methodology/REGISTRY.md:L1849-L1911. Concrete fix: None.
• Severity: P3. Impact: The absorb + fixed_effects changelog note aligns with the existing FWL guard in CHANGELOG.md:L27-L27, diff_diff/estimators.py:L263-L270, and diff_diff/estimators.py:L1028-L1035. Concrete fix: None.
• Severity: P3. Impact: The prior power-analysis wording issue is addressed. CHANGELOG.md:L22-L25 now scopes built-in support to registry-backed estimators, consistent with diff_diff/power.py:L446-L535 and diff_diff/power.py:L1419-L1463. Concrete fix: None.

Code Quality

• Severity: P3. Impact: Version metadata is internally consistent across diff_diff/init.py:L190-L190, pyproject.toml:L7-L7, rust/Cargo.toml:L3-L3, docs/llms-full.txt:L5-L5, and CHANGELOG.md:L954-L954. Concrete fix: None.

Performance

• Severity: P3. Impact: No runtime-path or algorithmic code changed; the diff is release metadata and changelog text only. Concrete fix: None.

Maintainability

• Severity: P3. Impact: The PR is narrow and coherent, and it does not introduce new maintenance risk beyond the existing manual multi-file version bump workflow. Concrete fix: None.

Tech Debt

• Severity: P3. Impact: This PR does not introduce a new deferred limitation that needs tracking in TODO.md:L44-L73. Concrete fix: None.

Security

• Severity: P3. Impact: No secrets, credentials, or attack-surface changes appear in the touched files. Concrete fix: None.

Documentation/Tests

• Severity: P3. Impact: The changelog’s EfficientDiD validation claim is supported by the existing committed validation suite and fixture provenance in tests/test_efficient_did_validation.py:L1-L230 and tests/data/README.md:L3-L54. Concrete fix: None.
• Severity: P3. Impact: No test changes are needed for the version-string updates themselves, and I did not find a new documentation mismatch in the touched files after the power-analysis wording correction. Concrete fix: None.