Release v3.5.2: version bump + CHANGELOG reconciliation

[igerber]

Summary

• Bump version 3.5.1 → 3.5.2 across diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, and CITATION.cff (date-released: 2026-06-08).
• Reconcile the CHANGELOG. The Firpo & Possebom (2018) confidence-sets-by-test-inversion feature (PR SyntheticControl: confidence sets by test inversion (Firpo-Possebom 2018, PR-B) #527) was filed under [3.5.1] but merged after the v3.5.1 tag was cut, so the tagged v3.5.1 did not actually contain it. That entry now lives in the new ## [3.5.2] section.
• [3.5.2] collects everything that landed post-tag: CBWSDID covariate balancing (Add CBWSDID covariate balancing to StackedDiD (Ustyuzhanin 2026) #534), SyntheticControl conformal inference (SyntheticControl: conformal inference (Chernozhukov-Wüthrich-Zhu 2021) #530), the Firpo-Possebom confidence sets (SyntheticControl: confidence sets by test inversion (Firpo-Possebom 2018, PR-B) #527), the placebo_effects → variance_effects rename + deprecation (refactor: rename SyntheticDiDResults.placebo_effects → variance_effects (deprecated alias) #532), and the ImputationDiD methodology validation + SE corrections (ImputationDiD methodology validation (PR-B): exact FE variance + unit-clustered Eq.8 + R parity #533).
• The Firpo PR-A paper review (docs: Firpo & Possebom (2018) paper review — SCM CI by test inversion (PR-A) #524, docs-only) stays in [3.5.1] — it was genuinely in that tag. A fresh empty [Unreleased] heading and the [3.5.2] comparison link were added.

Methodology references (required if estimator / math changes)

• Method name(s): N/A — no estimator/math code changes. The only diff_diff/ change is the __version__ string literal; all other changes are release metadata + CHANGELOG prose. The features being re-attributed (CBWSDID, conformal, Firpo-Possebom, ImputationDiD) were validated in their own prior PRs.
• Paper / source link(s): N/A (no methodology change in this PR)
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No test changes (release bump + changelog only)
• Backtest / simulation / notebook evidence (if applicable): N/A
• Verified all version strings are consistent at 3.5.2; diff_diff/__init__.py parses; CHANGELOG diff is a clean move of the misfiled SyntheticControl: confidence sets by test inversion (Firpo-Possebom 2018, PR-B) #527 bullet with no content lost.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

🤖 Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good. No unmitigated P0/P1 findings.

Executive Summary

• This PR is release metadata and changelog reconciliation only; no estimator, weighting, variance/SE, or default behavior changes are introduced.
• The moved Firpo-Possebom changelog entry matches the documented SyntheticControl methodology and documented deviations in REGISTRY.md.
• Version surfaces changed in the diff are consistent at 3.5.2.
• One minor stale internal comment still says “current is 3.5.1”; informational only.

Methodology

No P0/P1 findings.

The SyntheticControl Firpo-Possebom changelog prose now under 3.5.2 aligns with the registry’s documented contract: opt-in test inversion, strict p^f > gamma, analytical inference fields staying NaN, exact breakpoint inversion as a documented implementation choice, and no R anchor validation. See CHANGELOG.md:L10-L34 and docs/methodology/REGISTRY.md:L2021-L2053.

Code Quality

No findings. Only __version__ changed in package code at diff_diff/__init__.py:L301.

Performance

No findings. No runtime code paths changed.

Maintainability

• Severity: P3
• Impact: The version bump leaves a stale internal bump-time comment saying “current is 3.5.1”, which can confuse future release maintenance but does not affect runtime behavior.
• Concrete fix: Update or remove the comment at diff_diff/results.py:L1217-L1218.

Tech Debt

No blocking findings. The only noted issue is the P3 stale release-maintenance comment above.

Security

No findings. No secrets, credentials, network logic, or input-handling code changed.

Documentation/Tests

No blocking findings.

Version metadata is consistent across the changed release surfaces: CITATION.cff:L10-L11, diff_diff/__init__.py:L301, diff_diff/guides/llms-full.txt:L5, pyproject.toml:L7, and rust/Cargo.toml:L3.

Validation note: static version parsing passed. Import validation could not run because this review environment lacks numpy, so I did not treat that as a PR issue.

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 8652bb98f3b389c5ebb2475216d68b61dffccce0

---

Overall Assessment

✅ Looks good. No unmitigated P0/P1 findings.

Executive Summary

• Re-review focus: the prior P3 stale release comment in diff_diff/results.py is resolved.
• No estimator, weighting, variance/SE, inference, or default behavior code paths are changed.
• The moved Firpo-Possebom changelog entry remains aligned with the documented SyntheticControl registry notes.
• Version surfaces are consistent at 3.5.2.
• git diff --check passed; no whitespace/errors detected.

Methodology

No findings.

Affected method prose: SyntheticControl confidence sets by test inversion, changelog-only at CHANGELOG.md:L15. The registry documents the non-analytical confidence-set contract, strict p^f > gamma convention, exact breakpoint/grid construction choice, and no-R-anchor validation at docs/methodology/REGISTRY.md:L2021-L2053. No estimator math or inference implementation changed.

Code Quality

No findings.

Only touched package code is the version literal at diff_diff/__init__.py:L301 and the deprecation directive/comment cleanup at diff_diff/results.py:L1209-L1225.

Performance

No findings. No runtime performance paths changed.

Maintainability

No findings.

Re-review note: the previous stale “current is 3.5.1” comment was removed, and the deprecation directive now matches the release where the alias is introduced: diff_diff/results.py:L1210-L1216.

Tech Debt

No findings. No new deferred work is introduced.

Security

No findings. No secrets, credentials, network logic, or input-handling changes.

Documentation/Tests

No findings.

Version metadata is consistent across changed release surfaces: CITATION.cff:L10-L11, pyproject.toml:L7, rust/Cargo.toml:L3, diff_diff/__init__.py:L301, and diff_diff/guides/llms-full.txt:L5. The new changelog section and comparison link are present at CHANGELOG.md:L10 and CHANGELOG.md:L1588.