Release v3.0.2

[igerber]

Summary

• Bump version to 3.0.2 across all locations (pyproject.toml, init.py, Cargo.toml, llms-full.txt, CITATION.cff)
• Finalize CHANGELOG entry for v3.0.2 (promoted from [Unreleased])

What's in v3.0.2

New estimator:

• ChaisemartinDHaultfoeuille (alias DCDH) - reversible-treatment DiD, Phases 1+2 (DID_M, multi-horizon event study, dynamic placebos, sup-t bands)

New features:

• Survey-aware power analysis (SurveyPowerConfig, expected_deff, simulation path with survey_design)
• aggregate_survey() pweight default (second_stage_weights parameter)
• conditional_pt parameter on survey DGP
• Tutorial 18 (geo-experiments with SyntheticDiD)
• Practitioner decision tree and getting started guide (B1b-d)

Infrastructure:

• Standalone CI Gate workflow (doc-only PRs no longer block)
• JOSS paper (paper.md, paper.bib)
• CONTRIBUTORS.md, CITATION.cff with author credit

Methodology references (required if estimator / math changes)

• Method name(s): Version bump only - all methodology changes were reviewed in their respective PRs (Add practitioner decision tree and getting started guide (B1b-d) #287-feat: add second_stage_weights parameter to aggregate_survey() #299)
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No test changes (version metadata only)
• All methodology PRs (Add ChaisemartinDHaultfoeuille (dCDH) DID_M estimator (Phase 1) #290, feat: survey-aware power analysis (SurveyPowerConfig + deff) #292, Add Phase 2 multi-horizon event study for dCDH estimator #294, feat: add second_stage_weights parameter to aggregate_survey() #299) passed CI individually

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 metadata/documentation-only. I did not find any code-level changes to estimators, weighting, variance/SE logic, identification checks, or runtime defaults.
• Cross-checking the methodology-relevant release-note items against the Methodology Registry found no new undocumented methodology deviations in this PR.
• Version metadata is internally consistent across the touched package/runtime files: pyproject.toml:L7, diff_diff/init.py:L231, rust/Cargo.toml:L3, CITATION.cff:L10-L11, and docs/llms-full.txt:L5.
• The only actionable issues are documentation accuracy problems in the promoted 3.0.2 changelog: one misstates the survey-power API/support, and one overstates aggregate_survey() compatibility.
• docs/llms-full.txt still sits outside doc-snippet CI coverage, but that gap is already tracked in TODO.md:L86-L94.

Methodology

• No findings. The affected methods referenced in the release notes are already documented in the registry, including dCDH scope/deviations, survey power, aggregate_survey() weighting semantics, and conditional_pt: REGISTRY.md:L460-L628, REGISTRY.md:L2287-L2288, REGISTRY.md:L2478-L2515, REGISTRY.md:L2753-L2764. No P0/P1 methodology issue is introduced by this diff.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• Severity P3 | Impact: This PR updates docs/llms-full.txt:L5, but .txt AI guides are still outside snippet-smoke validation, so future API/doc drift in that surface can slip through CI. | Concrete fix: No action required for this PR; this is already tracked in TODO.md:L93-L94.

Security

• No findings. The diff is release metadata/docs only, and I did not see any secrets or sensitive material.

Documentation/Tests

• Severity P2 | Impact: CHANGELOG.md:L18 documents a non-existent survey-power API: it says SurveyPowerConfig has expected_deff, that simulation power accepts survey_design, and that this works “with any estimator.” The actual public API uses deff for the analytical helpers, survey_config for simulate_power/simulate_mde/simulate_sample_size, and the survey-DGP path only supports a subset of estimators. Users following the release note will hit wrong signatures or unsupported-estimator errors. | Concrete fix: Rewrite the bullet to match diff_diff/power.py:L57-L125, diff_diff/power.py:L1767-L1785, diff_diff/power.py:L1834-L1840, diff_diff/power.py:L3026-L3167, and REGISTRY.md:L2287-L2288: use deff for the analytical helpers, survey_config for the simulation helpers, and SurveyPowerConfig(survey_design=...) for custom design override.
• Severity P2 | Impact: CHANGELOG.md:L19 says aggregate_survey() pweight output is compatible with “all 16 estimators.” The project’s own support matrix says the default pweight output is compatible with all survey-capable estimators, not all estimators; ChaisemartinDHaultfoeuille has no survey support and BaconDecomposition is diagnostic only. This also contradicts the nearby “docs no longer overclaim universal estimator compatibility” note at CHANGELOG.md:L36. | Concrete fix: Change the wording to “all survey-capable estimators” and reserve aweight for the estimators marked Full in diff_diff/prep.py:L1468-L1482, REGISTRY.md:L2494-L2512, docs/choosing_estimator.rst:L691-L696, and docs/choosing_estimator.rst:L723-L800.
• Severity P3 | Impact: CHANGELOG.md:L31 says CITATION.cff was updated with the author name, but this diff only changes version/date; the author entry was already present in CITATION.cff:L5-L11. | Concrete fix: Narrow that changelog bullet to the actual metadata change, e.g. version/date only.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: a6b1a8376d0956195ec6ae9f32f2d4ba8419ef88

---

Overall Assessment

✅ Looks good

Executive Summary

• This remains a release-metadata / release-notes PR. I did not find any code changes to estimator logic, weighting, variance/SE computation, inference paths, or identification checks.
• The prior non-blocking changelog issues are resolved: the survey-power bullet now matches the shipped deff / survey_config API, the aggregate_survey() bullet now matches the survey support matrix, and the CITATION.cff note is now limited to version/date updates. See CHANGELOG.md:L18, CHANGELOG.md:L19, CHANGELOG.md:L31, docs/methodology/REGISTRY.md:L2287, docs/methodology/REGISTRY.md:L2495, and CITATION.cff:L10.
• Version metadata is internally consistent across pyproject.toml:L7, diff_diff/init.py:L231, rust/Cargo.toml:L3, CITATION.cff:L10, docs/llms-full.txt:L5, and CHANGELOG.md:L10.
• No methodology defect is introduced by the release-note text: the 3.0.2 entries for ChaisemartinDHaultfoeuille, survey-aware power, aggregate_survey(), and conditional_pt align with the Methodology Registry and public docstrings/API. See docs/methodology/REGISTRY.md:L460, docs/methodology/REGISTRY.md:L2288, docs/methodology/REGISTRY.md:L2478, docs/methodology/REGISTRY.md:L2753, diff_diff/chaisemartin_dhaultfoeuille.py:L246, diff_diff/power.py:L58, diff_diff/prep.py:L1419, and diff_diff/prep_dgp.py:L1163.
• Severity P2 [Newly identified]: the promoted release notes say the README received aggregate_survey() documentation updates, but the top-level README still describes the old aweight default. See CHANGELOG.md:L29, CHANGELOG.md:L30, and README.md:L85.
• Severity P3: docs/llms-full.txt:L5 is still outside doc-snippet CI coverage, but that limitation is already tracked in TODO.md:L93.

Methodology

• No findings. The methodology-adjacent changelog bullets now match the registry and in-code documentation for dCDH scope, survey-power deff / survey_config, aggregate_survey() weighting semantics, and conditional_pt. No runtime methodology change is introduced in this diff.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• Severity P3 | Impact: docs/llms-full.txt:L5 is touched in this PR, but .txt AI-guide docs are still outside the doc-snippet smoke tests tracked in TODO.md:L93. Future drift on that surface can still bypass CI. | Concrete fix: No action required for this PR; this is already properly tracked in TODO.md.

Security

• No findings.

Documentation/Tests

• Severity P2 [Newly identified] | Impact: CHANGELOG.md:L29 says aggregate_survey() now defaults to pweight, and CHANGELOG.md:L30 says the README was updated accordingly, but README.md:L85 still tells users the returned second-stage design uses aweight by default. That leaves 3.0.2 with conflicting user-facing documentation about default weighting and downstream estimator compatibility. | Concrete fix: Update README.md:L85 to match diff_diff/prep.py:L1427, docs/methodology/REGISTRY.md:L2495, and docs/choosing_estimator.rst:L692: default pweight, with second_stage_weights="aweight" as an opt-in mode for estimators marked Full.