Release 3.3.2: dCDH by_path × trends extensions, Yatchew mean_independence, HAD Phase 4 R-parity

[igerber]

Summary

• Bump version 3.3.1 → 3.3.2 across the 6 version surfaces
• Rename CHANGELOG `[Unreleased]` → `[3.3.2] - 2026-04-26`; add comparison link

Highlights since 3.3.1

• dCDH `by_path` × `trends_linear` / `trends_nonparam` (PR dCDH by_path: lift trends_linear + trends_nonparam gates (Wave 3 #6+#7) #393) — per-path cumulated DID^{fd} surfaces, state-set trends, R-parity goldens. F_g=3 boundary divergence flagged with targeted UserWarning.
• dCDH `by_path` + `controls` (PR Lift by_path + controls gate (DID^X residualization) #378) — DID^X residualization compatible with by_path; multi-baseline UserWarning when paths share heterogeneous F_g.
• `yatchew_hr_test(null="mean_independence")` (PR Add yatchew_hr_test(null='mean_independence') mode #397) — closes the placebo Yatchew R-parity gap (R `order=0` mode).
• HAD Phase 4: `trends_lin` (Eq 17/18) + R end-to-end parity (PR HAD Phase 4: trends_lin (Eq 17/18) + R-package end-to-end parity #392) — new `benchmarks/data/did_had_golden.json` cross-language fixture and full-pipeline parity tests.
• Rust `rand` 0.8 → 0.10, `rand_xoshiro` 0.6 → 0.8 (PR chore(rust): bump rand 0.8 -> 0.10, rand_xoshiro 0.6 -> 0.8 #391) — MSRV bump + bit-identity snapshot test guards bootstrap RNG output.
• Tutorial 20: HAD for National Brand Campaign (PR Add Tutorial 20: HAD for National Brand Campaign with Regional Spend Intensity #394) + HAD LinkedIn carousel (PR Add LinkedIn carousel for HAD estimator announcement #398).
• README + docs: cumulative downloads badge swap (PR README: swap monthly downloads badge for cumulative; drop docs-passing badge #395), drop docs-passing badge, community standards files (CODE_OF_CONDUCT, SECURITY, issue templates) + BRIEFING.md cleanup (PR chore: add community standards files; remove stray BRIEFING.md #388), `docs-tests.yml` workflow carved out from rust-test (PR Add docs-tests.yml; remove test_doc_snippets.py from rust-test.yml #399).
• HAD doc-snippet bug fixes (PR Fix latent doc-snippet bugs from PR #389 (HAD ecosystem) #396).

Methodology references

• N/A for this PR — version-bump only; no source code, methodology, or Rust changes
• Methodology citations for all underlying work are in `docs/methodology/REGISTRY.md` and the section-level CHANGELOG entries

Validation

• No test changes (release-bump PR)
• All methodology + correctness validation lives in the underlying merged PRs (Lift by_path + controls gate (DID^X residualization) #378, chore: add community standards files; remove stray BRIEFING.md #388, chore(rust): bump rand 0.8 -> 0.10, rand_xoshiro 0.6 -> 0.8 #391, HAD Phase 4: trends_lin (Eq 17/18) + R-package end-to-end parity #392, dCDH by_path: lift trends_linear + trends_nonparam gates (Wave 3 #6+#7) #393, Add Tutorial 20: HAD for National Brand Campaign with Regional Spend Intensity #394, README: swap monthly downloads badge for cumulative; drop docs-passing badge #395, Fix latent doc-snippet bugs from PR #389 (HAD ecosystem) #396, Add yatchew_hr_test(null='mean_independence') mode #397, Add LinkedIn carousel for HAD estimator announcement #398, Add docs-tests.yml; remove test_doc_snippets.py from rust-test.yml #399 — see CHANGELOG `[3.3.2]` for the full list)

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 release metadata only: version bumps plus changelog promotion from Unreleased to 3.3.2; I found no in-scope estimator, weighting, variance/SE, identification, or default-behavior code changes.
• The methodology-heavy changelog content already maps to documented registry notes/deviations for dCDH by_path, HAD trends_lin, and yatchew_hr_test(null="mean_independence"), so there is no new undocumented methodology mismatch in scope (docs/methodology/REGISTRY.md:L641-L675, docs/methodology/REGISTRY.md:L2470-L2503, TODO.md:L104-L106).
• The main release/version surfaces are internally synchronized to 3.3.2 in CHANGELOG.md:L8, CITATION.cff:L10-L11, diff_diff/__init__.py:L290, diff_diff/guides/llms-full.txt:L5, pyproject.toml:L7, and rust/Cargo.toml:L3.
• P2: as written, 3.3.2 is carrying additive public features/APIs, which conflicts with the project’s own SemVer statement in CHANGELOG.md:L5-L6.
• P3: one ancillary release-collateral script outside the diff still hardcodes v3.3.1, so regenerated marketing output would be stale (carousel/generate_had_carousel.py:L173).

Methodology

• Severity: None. Impact: No estimator implementation, math, SE/inference logic, assumption checks, or defaults changed in the diff. The items described in CHANGELOG.md:L11-L14 are already documented in the Methodology Registry, including explicit notes/deviations for dCDH by_path and HAD/Yatchew parity behavior (docs/methodology/REGISTRY.md:L641-L675, docs/methodology/REGISTRY.md:L2470-L2503), and the HAD trends_lin × survey_design limitation is already tracked in TODO.md:L105. Concrete fix: None.

Code Quality

• Severity: None. Impact: The edited files are simple metadata/doc surfaces; no new executable logic or error-handling paths were introduced. Concrete fix: None.

Performance

• Severity: None. Impact: Metadata and changelog edits do not affect Python or Rust runtime behavior. Concrete fix: None.

Maintainability

• Severity: P2. Impact: CHANGELOG.md says the project adheres to Semantic Versioning (CHANGELOG.md:L5-L6), but the 3.3.2 entry advertises additive public functionality such as new by_path compatibility/results fields, yatchew_hr_test(null="mean_independence"), and HAD trends_lin (CHANGELOG.md:L11-L13). That makes a patch release behave like a minor release for downstream consumers who rely on patch-version stability. Concrete fix: Either publish this bundle as 3.4.0 and update the version surfaces accordingly, or narrow the 3.3.2 changelog entry to true patch-level changes only.

Tech Debt

• Severity: None. Impact: The deferred items mentioned by the release notes are already tracked, so this PR does not introduce new untracked technical debt. Concrete fix: None.

Security

• Severity: None. Impact: No secrets, credentials, or executable security-sensitive code paths are introduced by the diff. Concrete fix: None.

Documentation/Tests

• Severity: P3. Impact: Outside the edited files, carousel/generate_had_carousel.py:L173 still contains v3.3.1; if that script is rerun for this release, it will emit stale collateral despite the package/release being 3.3.2. This is informational only and does not affect library correctness. Concrete fix: Update the literal to v3.3.2 or source the displayed version from a single canonical version constant.