Release 3.2.0: BusinessReport preview, dCDH survey completion, silent-failures audit

[igerber]

Summary

• Bump version to 3.2.0 across diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, and CITATION.cff. Per project SemVer convention, minor bumps are reserved for new estimators or new module-level public API — PR Add BusinessReport and DiagnosticReport (experimental preview) #318 adds BusinessReport, DiagnosticReport, and DiagnosticReportResults to diff_diff/__init__.py, driving the minor bump.
• Populate CHANGELOG [3.2.0] entry covering 18 merged PRs since v3.1.3:
  • Added (3): BusinessReport + DiagnosticReport experimental preview (Add BusinessReport and DiagnosticReport (experimental preview) #318), kernel/local-linear/nprobust foundation for did-no-untreated (Phase 1a: Kernel infrastructure + HC2/Bell-McCaffrey variance #327, Phase 1b: MSE-optimal bandwidth selector (CCF 2018 DPI) #335), dCDH cell-period IF allocator Class A contract (Add cell-period IF allocator for dCDH survey variance #323)
  • Changed (5): Zenodo DOI badge + CITATION.cff identifiers (Add Zenodo DOI badge, citation metadata, and identifiers #325), dCDH heterogeneity + within-group-varying PSU under Binder TSL (Extend dCDH heterogeneity SE to cell-period allocator #329), dCDH bootstrap within-group-varying PSU (Extend dCDH PSU-level wild bootstrap to cell granularity #332), performance-review scenarios doc + benchmark harness extension (Add practitioner-workflow performance baseline #333), flaky wall-clock timing tests excluded from CI (Exclude flaky wall-clock timing tests from default CI #330, Skip wall-clock timing tests on CI to stop false-positive flake #336)
  • Fixed (6): silent-failures audit closeouts on axes A/C/J/E/G (Surface silent np.linalg.solve fallbacks across axis-A minor solver paths #334, Close axis-C/J silent-failures audit: B-spline derivative + PA survey cache #339, Surface row-count for axis-E silent coercion / drop sites #331, Add axis-G Rust vs Python backend parity edge-case tests #337), SDID diagnostic Y-normalization parity (Extend PR #312 Y-normalization contract into SDID diagnostic methods #328), TROP bootstrap failure-rate guards (Guard TROP bootstrap loops against silent high-failure-rate runs #324), simulate_power() failure-count surface + narrow except clause (Surface PowerAnalysis simulation-failure count and narrow except clause #326)
• BusinessReport and DiagnosticReport ship as experimental preview per the project's existing BR/DR positioning — wording, verdict thresholds, and schema shape will continue to evolve. Downstream tooling should not anchor on the current schema yet.
• CITATION.cff retains the v3.1.3 versioned DOI in identifiers:; the v3.2.0 versioned DOI gets minted by Zenodo when the GitHub Release is cut, and a follow-up PR can add it to the list.

Methodology references

• Method name(s): N/A — this is release metadata; no estimator, math, or inference changes in this PR. The underlying methodology for each included PR is documented in its own merged PR and in docs/methodology/REGISTRY.md / docs/methodology/REPORTING.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 18 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 3.2.0 / date-released: 2026-04-19).

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment

✅ Looks good

Executive Summary

• This is a release-metadata PR in executable terms: the only code change is the version constant bump in diff_diff/init.py:L255.
• I cross-checked the methodology-heavy changelog additions in CHANGELOG.md:L10-L31 against the documented reporting-layer notes in docs/methodology/REPORTING.md:L58-L177 and the dCDH survey/bootstrap registry notes in docs/methodology/REGISTRY.md:L622-L627 and docs/methodology/REGISTRY.md:L669-L670; I did not find any undocumented methodology mismatch.
• Version surfaces are aligned across diff_diff/init.py:L255, pyproject.toml:L7, rust/Cargo.toml:L3, diff_diff/guides/llms-full.txt:L5, CITATION.cff:L10-L19, and CHANGELOG.md:L10-L31.
• The stated minor bump is consistent with the current public API surface: BusinessReport, DiagnosticReport, and DiagnosticReportResults are exported from diff_diff/init.py:L216-L225 and diff_diff/init.py:L440-L445.
• No unmitigated P0 or P1 findings.

Methodology

• Severity: none. Impact: none. Concrete fix: none. The diff does not modify estimator logic, weighting, variance/SE computation, identification checks, or defaults. The new release-note text is consistent with the documented reporting deviations and dCDH survey/bootstrap contracts in docs/methodology/REPORTING.md:L17-L28, docs/methodology/REPORTING.md:L58-L177, docs/methodology/REGISTRY.md:L622-L627, and docs/methodology/REGISTRY.md:L669-L670.

Code Quality

• Severity: none. Impact: none. Concrete fix: none. No new logic or code path is introduced; the changed Python code is limited to the package version constant in diff_diff/init.py:L255. The edge-case checklist is not implicated because there is no new estimator/inference path to exercise.

Performance

• Severity: none. Impact: none. Concrete fix: none. Metadata and changelog edits do not affect runtime behavior.

Maintainability

• Severity: none. Impact: none. Concrete fix: none. Version propagation is coherent across the touched surfaces, and docs already derive their displayed version from diff_diff.__version__ in docs/conf.py:L19-L20, which helps prevent future drift.

Tech Debt

• Severity: none. Impact: none. Concrete fix: none. I did not see any new deferred correctness limitation or newly introduced TODO-worthy gap in the changed files.

Security

• Severity: none. Impact: none. Concrete fix: none. No secrets, credentials, dependency changes, executable-surface expansion, or unsafe input-handling changes appear in this diff.

Documentation/Tests

• Severity: P3. Impact: CITATION.cff now advertises version 3.2.0, but its version-specific DOI list still stops at v3.1.3; tooling that expects a release-specific DOI directly from the CFF may not find one for 3.2.0 yet. Concrete fix: after Zenodo mints the 3.2.0 DOI, add it to identifiers: and optionally make it the top-level doi if version-specific citation is preferred. CITATION.cff:L10-L19, README.md:L3111-L3115
• Severity: none. Impact: none. Concrete fix: none. No new tests are needed for this PR because it does not change executable behavior.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 852953b61b44f2d8e541ffd7e968f4f262e68b81

---

Overall Assessment

✅ Looks good

This re-review resolves the prior informational DOI concern. I found no unmitigated P0 or P1 issues in the changed files.

Executive Summary

• The executable surface is still metadata-only: the only code change is the version constant bump in diff_diff/init.py:L255-L255, with matching version updates in pyproject.toml:L5-L8, rust/Cargo.toml:L1-L4, CITATION.cff:L10-L14, and diff_diff/guides/llms-full.txt:L1-L5.
• The methodology-heavy 3.2.0 changelog bullets are consistent with the documented reporting and dCDH registry notes; I did not find an undocumented methodology deviation in the release text. CHANGELOG.md:L13-L20, docs/methodology/REPORTING.md:L58-L65, docs/methodology/REPORTING.md:L144-L197, docs/methodology/REGISTRY.md:L622-L627, docs/methodology/REGISTRY.md:L668-L670
• The prior CITATION.cff P3 is effectively addressed: the project now consistently documents a concept-DOI-only policy across the changelog, CFF, and README. CHANGELOG.md:L18-L18, CITATION.cff:L10-L14, README.md:L3111-L3113
• One minor documentation clarity issue remains: the new performance bullet says baselines were refreshed against “current main,” but the committed benchmark artifacts still identify diff_diff_version as 3.1.3. CHANGELOG.md:L21-L22, docs/performance-plan.md:L7-L22, benchmarks/speed_review/baselines/geo_few_markets_small_python.json:L53-L53, benchmarks/speed_review/baselines/reversible_dcdh_python.json:L43-L43

Methodology

• Severity: P3-informational. Impact: none. Concrete fix: none required. The dCDH survey cell-period allocator referenced in the changelog is a documented library convention rather than a paper-derived observation-level linearization, but that deviation is explicitly disclosed in the registry and already tracked in TODO, so it is mitigated under the review rules. CHANGELOG.md:L15-L20, docs/methodology/REGISTRY.md:L668-L670, TODO.md:L55-L60
• Severity: none. Impact: none. Concrete fix: none. I found no undocumented mismatch between the new changelog claims and the documented reporting-layer notes on no-traffic-light gates, pre-trends thresholds, and power-aware phrasing, or the dCDH survey/bootstrap notes. CHANGELOG.md:L13-L20, docs/methodology/REPORTING.md:L58-L65, docs/methodology/REPORTING.md:L144-L197, docs/methodology/REGISTRY.md:L622-L627, docs/methodology/REGISTRY.md:L635-L635, docs/methodology/REGISTRY.md:L668-L670

Code Quality

• Severity: none. Impact: none. Concrete fix: none. No new executable logic or inference path is introduced here; the Python code change is limited to the package version constant. diff_diff/init.py:L255-L255

Performance

• Severity: P3. Impact: the release note can read as if the committed benchmark baselines were captured on 3.2.0/current main, while the checked-in artifacts and findings doc still label that measurement surface as v3.1.3. Concrete fix: change the changelog wording to “refreshed against then-current main (v3.1.3)” or refresh the baseline metadata/docs if 3.2.0 measurements were intended. CHANGELOG.md:L21-L22, docs/performance-plan.md:L7-L22, benchmarks/speed_review/README.md:L77-L79, benchmarks/speed_review/baselines/geo_few_markets_small_python.json:L53-L53, benchmarks/speed_review/baselines/reversible_dcdh_python.json:L43-L43
• Severity: none. Impact: none. Concrete fix: none. The actual version bumps themselves do not affect runtime behavior. diff_diff/init.py:L255-L255, pyproject.toml:L5-L8, rust/Cargo.toml:L1-L4

Maintainability

• Severity: none. Impact: none. Concrete fix: none. The canonical version surfaces are synchronized, and the minor bump is coherent with the current top-level API already exporting BusinessReport, DiagnosticReport, and DiagnosticReportResults. diff_diff/init.py:L216-L224, diff_diff/init.py:L255-L255, diff_diff/init.py:L440-L445, pyproject.toml:L5-L8, rust/Cargo.toml:L1-L4, CITATION.cff:L10-L14

Tech Debt

• Severity: none. Impact: none. Concrete fix: none. No new deferred correctness item is introduced by this PR; the only methodology-adjacent caveat implicated by the new changelog text is already tracked in TODO.md, so it is mitigated rather than a blocker. TODO.md:L55-L60, docs/methodology/REGISTRY.md:L668-L670

Security

• Severity: none. Impact: none. Concrete fix: none. I did not find any new secret exposure, dependency-surface change, or executable attack-surface expansion in this metadata-only diff. CITATION.cff:L1-L30, pyproject.toml:L1-L16, rust/Cargo.toml:L1-L10

Documentation/Tests

• Severity: none. Impact: none. Concrete fix: none. No new tests are needed for version-string and release-note edits because the diff does not change runtime behavior. The prior informational concern around CITATION.cff DOI handling is resolved by the now-explicit concept-DOI-only policy in the changelog/CFF/README. CHANGELOG.md:L18-L18, CITATION.cff:L10-L14, README.md:L3111-L3113