Release v3.5.3

[igerber]

Summary

• Bump version 3.5.2 → 3.5.3 across diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, and CITATION.cff (with date-released: 2026-06-25).
• Convert the curated ## [Unreleased] CHANGELOG section to ## [3.5.3] - 2026-06-25 and add the v3.5.2...v3.5.3 comparison link. No content regenerated — the existing hand-written release notes are preserved.

Changes shipped in this release (already merged to main since 3.5.2):

• Wild cluster bootstrap now imposes the null with test-inversion CIs, fixing an invalid p-value that contradicted its own CI ([Bug]: Result of Wild Cluster Bootstrap is inconsistent to Cluster-robust #543 / Fix wild cluster bootstrap p-value / CI inconsistency (WCR + test inversion, #543) #546); validated against R fwildclusterboot.
• TwoStageDiD methodology validation (Gardner 2022 / did2s) with exact GMM variance and R parity (TwoStageDiD methodology validation (PR-B): did2s parity + exact GMM variance #545).
• generate_synthetic_control_data() + tutorial 25 capstone SyntheticControl walkthrough (Add synthetic control tutorial + generate_synthetic_control_data generator #540).
• TripleDifference power now routes to the panel DGP when n_periods > 2 (fix: route TripleDifference power to panel DGP when n_periods > 2 #544).
• Degenerate-design SE guards (saturated-design ZeroDivision, rank-deficient vcov, tiny-negative-variance NaN SE).

Methodology references (required if estimator / math changes)

• Method name(s): N/A — version-string + CHANGELOG only. No source/math changes in this PR (the methodology work itself landed in the PRs listed above).
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No test changes — this is a release-metadata bump only.
• Backtest / simulation / notebook evidence (if applicable): N/A
• Version strings verified consistent across all six locations; no stale 3.5.2 references remain.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good

Executive Summary

• The PR is a release metadata cut from 3.5.2 to 3.5.3.
• No estimator, math, weighting, variance/SE, identification-assumption, or default-behavior code paths are changed.
• Version surfaces are consistent across pyproject.toml, diff_diff/__init__.py, rust/Cargo.toml, CITATION.cff, and llms-full.txt.
• Remaining 3.5.2 references are historical changelog/deprecation references, not stale release metadata.
• No P0/P1 findings.

Methodology

No findings.

• Severity: N/A
• Impact: No affected methods. The diff only promotes CHANGELOG.md from [Unreleased] to [3.5.3] - 2026-06-25 and bumps metadata versions. No methodology cross-check defect is introduced.
• Concrete fix: None.

Code Quality

No findings.

• Severity: N/A
• Impact: Metadata-only changes are straightforward and internally consistent.
• Concrete fix: None.

Performance

No findings.

• Severity: N/A
• Impact: No runtime code changed.
• Concrete fix: None.

Maintainability

No findings.

• Severity: N/A
• Impact: Release metadata is updated consistently. The changelog compare link exists at CHANGELOG.md:L1612.
• Concrete fix: None.

Tech Debt

No findings.

• Severity: N/A
• Impact: No new deferred work is introduced. Existing TODO.md tech-debt entries are unrelated to this release metadata bump.
• Concrete fix: None.

Security

No findings.

• Severity: N/A
• Impact: No secrets, credentials, or security-sensitive code changes observed in the diff.
• Concrete fix: None.

Documentation/Tests

No findings.

• Severity: P3 informational
• Impact: No tests were added, but this is acceptable for a metadata-only release PR with no behavioral code changes.
• Concrete fix: None required.