Bump version to 3.12.0

[igerber]

Summary

• Release bump 3.11.1 -> 3.12.0 (minor: two new public surfaces, one new kwarg, three behavioral corrections; nothing removed, no default flipped, so 4.0 stays reserved for the planned ledger rows).
• CHANGELOG.md: the ten changelog.d/ fragments since v3.11.1 compiled into ## [3.12.0] - 2026-09-08 by changelog_compile.py (fragments deleted; [Unreleased] stays pointer-only). Comparison link v3.11.1...v3.12.0 added.
• Version synced in diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, CITATION.cff (version + date-released: 2026-09-08).
• tests/test_naming_guard.py: (3, 12): "4.0" added to the release ladder (a missing key fails loudly; same shape as the 3.10.0 bump).

Release contents (12 PRs): DurationDiD estimator (#827), DMLDiD bad-control lane (#829) with tutorial 33 (#831), WooldridgeDiD(unsupported_period_action=) (#820), SDiD shape-only pre-fit diagnostic + placebo-anchored warning (#818), HC2/HC2-BM fail-closed at leverage one (#823), summary(alpha=) guard + honest dose-response bands (#811), exact fractional confidence-level labels (#815), three paper reviews (#810, #819, #822), ruff pin bump (#813).

Methodology references (required if estimator / math changes)

• Method name(s): N/A - version bump only; no estimator or math changes in this diff
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: tests/test_naming_guard.py (release-ladder entry only). Ran test_naming_guard.py + test_v4_matrix.py on Python 3.14 (CI's interpreter for that job) against this tree: 290 passed, 24 skipped; test_changelog_fragments.py, test_tracking_files.py, test_docs_ia.py pass; changelog_compile.py check OK; import diff_diff reports 3.12.0; no stale 3.11.1 references outside CHANGELOG history.
• Backtest / simulation / notebook evidence (if applicable): N/A

Security / privacy

• Confirm no secrets/PII in this PR: Yes

Changelog

• changelog.d/ fragment added (or N/A - no user-visible change): N/A - this PR compiles the existing fragments into the 3.12.0 section

[github-actions]

Overall assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive summary

• Release-only diff; no estimator, weighting, variance, identification, or default logic changed.
• Version metadata is consistently 3.12.0, with matching release dates and comparison link.
• Changelog compilation passes its repository check and preserves the pointer-only Unreleased section.
• Methodological claims align with documented Registry notes and tracked deviations.
• One P2 documentation inconsistency remains: several SDiD references name the unreleased version 3.11.2.

Methodology

• Severity: P3 — Informational.
  Impact: DMLDiD’s CS-simple weighting and DurationDiD’s PH/fixed-anchor choices differ from paper alternatives, but are explicitly documented and tracked; they are not defects. See docs/methodology/REGISTRY.md:L3406-L3420, docs/methodology/REGISTRY.md:L5983-L6002, and TODO.md:L60-L64.
  Concrete fix: None required.

No undocumented methodology, inference, or assumption-check changes were introduced.

Code Quality

No findings.

Performance

No findings; no executable performance paths changed.

Maintainability

No findings. The (3, 12): "4.0" release-ladder entry correctly preserves the enforcement horizon at tests/test_naming_guard.py:L103-L117.

Tech Debt

No untracked debt introduced. Limitations mentioned in the compiled notes remain recorded in TODO.md or DEFERRED.md.

Security

No findings or apparent secrets/PII.

Documentation/Tests

• Severity: P2.
  Impact: The release is identified as 3.12.0 at CHANGELOG.md:L12, but the SDiD methodology and compatibility comments say the diagnostic/schema shipped in nonexistent version 3.11.2. This gives users an inaccurate pickle-compatibility boundary. See docs/methodology/REGISTRY.md:L3511, diff_diff/results.py:L1287, and tests/test_methodology_sdid.py:L2590, L2655, L2761.
  Concrete fix: Replace the v3.11.2/pre-v3.11.2 references with v3.12.0/pre-v3.12.0, including affected test names or docstrings.

The changelog checker and metadata consistency checks pass. The full pytest suite could not be run because pytest is unavailable in the review environment.

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: cc85988e1617da203ef8b2665dcc99a2c8ded584

---

Overall assessment

✅ Looks good — no unmitigated P0/P1 findings.

Executive summary

• Prior 3.11.2 documentation inconsistency is fully resolved as 3.12.0.
• No estimator, inference, weighting, variance, or identification logic changed.
• All version metadata and release dates are consistent.
• Changelog compilation and comparison-link validation pass.

Methodology

• Severity: P3 — Informational.
  Impact: The SDiD diagnostic remains explicitly documented under a Registry Note; this PR only corrects its release boundary at docs/methodology/REGISTRY.md:L3511. Runtime methodology is unchanged.
  Concrete fix: None required.

Code Quality

No findings.

Performance

No findings; no performance-sensitive code changed.

Maintainability

No findings. The 3.12 lifecycle entry correctly retains the 4.0 enforcement horizon at tests/test_naming_guard.py:L110-L117.

Tech Debt

No new or untracked technical debt.

Security

No findings or apparent secrets/PII.

Documentation/Tests

The previous P2 finding is resolved consistently in diff_diff/results.py:L1287, docs/methodology/REGISTRY.md:L3511, and the affected SDiD tests.

Version surfaces agree at diff_diff/__init__.py:L376, pyproject.toml:L7, rust/Cargo.toml:L3, and CITATION.cff:L10-L11. The compiled release and comparison link validate at CHANGELOG.md:L12 and CHANGELOG.md:L5833.

The changelog compiler passes. Pytest could not run because pytest, numpy, and project dependencies are unavailable in the review environment.