Release 3.1.3: dCDH replicate weights, sparse fallback signaling, Zenodo config

[igerber]

Summary

• Bump version to 3.1.3 across diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, and CITATION.cff (first use of the updated /bump-version skill that includes CITATION.cff in scope).
• Populate CHANGELOG [3.1.3] entry covering five merged PRs:
  • Add replicate-weight variance and PSU-level bootstrap to dCDH #311 — Replicate-weight variance (variance_method="replicate" for BRR / Fay / JK1 / JKn / SDR) and PSU-level multiplier bootstrap for ChaisemartinDHaultfoeuille, with df-aware inference propagated through delta / HonestDiD surfaces.
  • Add Zenodo DOI auto-minting config for GitHub Releases #321 — .zenodo.json + LICENSE for Zenodo DOI auto-minting on future GitHub Releases.
  • Signal silent sparse -> dense lstsq fallback in ImputationDiD and TwoStageDiD #319 — ConvergenceWarning when ImputationDiD / TwoStageDiD silently fall back from sparse to dense lstsq.
  • Signal non-convergence in TROP alternating-minimization solvers #317 — ConvergenceWarning in TROP alternating-minimization solvers (numpy path, LOOCV, bootstrap aggregation) with top-level warning aggregation.
  • Add CITATION.cff to bump-version skill #320 — /bump-version skill now updates CITATION.cff and resolves a single RELEASE_DATE to prevent CHANGELOG/CITATION drift.
• Per project SemVer convention, this is a patch bump: additive extensions to existing estimators (PR Add replicate-weight variance and PSU-level bootstrap to dCDH #311's new variance_method values) are patch-level; minor bumps are reserved for new estimators or new module-level API.

Methodology references

• Method name(s): N/A for this release metadata PR. The underlying methodology for each shipped PR (dCDH replicate-weight variance, TROP convergence, etc.) is documented in its own merged PR and in docs/methodology/REGISTRY.md.
• Paper / source link(s): N/A (no methodology change in this PR itself).
• Any intentional deviations from the source: None.

Validation

• Per-PR regression coverage ships in the five merged PRs included in this release.
• No new tests in this PR — version-string and CHANGELOG edits only.
• Pre-release smoke: version strings consistent across the six canonical locations (confirmed by grep — all read 3.1.3 / date-released: 2026-04-18).

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good

Executive Summary

• P2: The new 3.1.3 changelog says ImputationDiD/TwoStageDiD sparse-fallback paths and TROP non-convergence emit ConvergenceWarning, but the current implementation, methodology docs, and tests still use UserWarning.
• No estimator/math/variance/default-behavior code is changed in this diff.
• The only methodology-related release note, dCDH replicate-weight variance / PSU bootstrap, is consistent with the Methodology Registry.
• The version/date bumps are internally consistent across the touched release surfaces, and CITATION.cff parses cleanly.
• The touched .txt guide remains outside doc-snippet CI, but that limitation is already tracked in TODO.md, so it is informational only.

Methodology

• No findings. The dCDH release note in CHANGELOG.md:L13 matches the documented survey/replicate-variance contract in docs/methodology/REGISTRY.md:L651-L653.

Code Quality

• No findings.

Performance

• No findings.

Maintainability

• No findings.

Tech Debt

• Severity: P3. Impact: This PR updates diff_diff/guides/llms-full.txt:L5, but .txt AI guides are still outside the current doc-snippet smoke tests, which is already tracked in TODO.md:L95. Concrete fix: None required for approval; keep the existing TODO item or add .txt coverage to tests/test_doc_snippets.py in a follow-up.

Security

• No findings.

Documentation/Tests

• Severity: P2. Impact: CHANGELOG.md:L17-L18 states that the shipped behavior is ConvergenceWarning, but the implementation still emits UserWarning in diff_diff/imputation.py:L1519-L1529, diff_diff/two_stage.py:L1655-L1666, and via the shared helper in diff_diff/utils.py:L68-L88. The current contract is also documented as UserWarning in docs/methodology/REGISTRY.md:L1082-L1086 and docs/methodology/REGISTRY.md:L1164-L1165, and asserted by tests/test_imputation.py:L888-L903, tests/test_two_stage.py:L494-L510, and tests/test_trop.py:L3983-L4026. This is a release-note accuracy issue that could mislead users filtering warnings by category. Concrete fix: Change the 3.1.3 changelog bullets to UserWarning, or rename the warning category in code and update the registry/tests in the same release.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 725c4529e9e9771c593bc5c6dbc6030f7501fe99

---

Overall Assessment

✅ Looks good

Executive Summary

• The prior re-review issue is resolved: the touched changelog entries now say UserWarning, which matches the current warning helper, registry text, and regression tests for sparse fallback and non-convergence signaling (CHANGELOG.md:L16-L18, CHANGELOG.md:L25-L28, diff_diff/utils.py:L75-L88, docs/methodology/REGISTRY.md:L1082-L1086, docs/methodology/REGISTRY.md:L1164-L1165, docs/methodology/REGISTRY.md:L1976).
• No estimator, math, weighting, variance, identification, or default-behavior code changes are introduced in this diff; it is release metadata plus changelog text (CHANGELOG.md:L10-L21, CITATION.cff:L10-L11, pyproject.toml:L7, diff_diff/init.py:L234, rust/Cargo.toml:L3).
• The new dCDH release note is consistent with the Methodology Registry’s documented replicate-weight variance and PSU-level bootstrap contract (CHANGELOG.md:L12-L13, docs/methodology/REGISTRY.md:L651-L653, docs/methodology/REGISTRY.md:L2665-L2713).
• Version surfaces are internally consistent at 3.1.3 across the touched files (pyproject.toml:L7, diff_diff/init.py:L234, rust/Cargo.toml:L3, CITATION.cff:L10-L11, diff_diff/guides/llms-full.txt:L5).
• Informational only: the touched .txt AI guide remains outside doc-snippet CI, but that gap is already tracked in TODO.md:L95.

Methodology

• No findings. The only methodology-facing addition is the 3.1.3 dCDH changelog summary, and it matches the registry’s documented support for BRR/Fay/JK1/JKn/SDR replicate variance, reduced survey df, HonestDiD propagation, and PSU-level bootstrap under survey designs (CHANGELOG.md:L12-L13, docs/methodology/REGISTRY.md:L651-L653, docs/methodology/REGISTRY.md:L2665-L2713).

Code Quality

• No findings. The prior warning-category mismatch in release text is fixed, and the touched release metadata is internally consistent (CHANGELOG.md:L16-L18, CHANGELOG.md:L25-L28, CITATION.cff:L10-L11, diff_diff/init.py:L234, pyproject.toml:L7, rust/Cargo.toml:L3, diff_diff/guides/llms-full.txt:L5).

Performance

• No findings. No runtime code paths or numerical kernels are changed in this diff.

Maintainability

• No findings. The new release entry is structurally complete and includes the corresponding compare link (CHANGELOG.md:L10-L21, CHANGELOG.md:L1322-L1323).

Tech Debt

• Severity: P3. Impact: diff_diff/guides/llms-full.txt:L5 is modified in this PR, but .txt AI guides still sit outside the current doc-snippet smoke tests, so drift on that surface would not be caught automatically. This is already tracked in TODO.md:L95. Concrete fix: No action required for approval; keep the existing TODO entry or extend tests/test_doc_snippets.py to validate .txt guides.

Security

• No findings. I did not see secrets, credentials, or risky config changes in the touched files.

Documentation/Tests

• No findings. The re-review target is addressed: changelog warning wording now matches the current implementation and tests for sparse-fallback and non-convergence behavior (CHANGELOG.md:L16-L18, CHANGELOG.md:L25-L28, diff_diff/imputation.py:L1518-L1529, diff_diff/two_stage.py:L1655-L1666, tests/test_imputation.py:L888-L904, tests/test_two_stage.py:L494-L511, tests/test_trop.py:L3983-L4027).