Bump version to 3.6.2

[igerber]

Summary

• Bump version 3.6.1 → 3.6.2 across all surfaces (diff_diff/__init__.py, pyproject.toml, rust/Cargo.toml, diff_diff/guides/llms-full.txt, CITATION.cff).
• Convert the curated ## [Unreleased] CHANGELOG section to ## [3.6.2] - 2026-07-03 and add the v3.6.1...v3.6.2 comparison link. No auto-generated content - the hand-written entries are preserved verbatim.

This release cuts the work merged to main since v3.6.1:

• FE-absorption MAP-demeaning overhaul: factorize-once + np.bincount engine, max_iter 100->10k parity, FE-spanned regressor guard, Rust demean_map kernel, opt-in chunked dispatch, and benchmark lane (bench: FE-absorption benchmark lane with BEFORE baselines + pyfixest yardstick (scenarios 7-13) #600, perf: factorize-once + bincount MAP demeaning, max_iter 10k parity, FE-spanned regressor guard #601, perf(rust): compiled MAP demeaning kernel - rayon across columns, 1.4-2.8x on FE-absorption workloads #603, perf(utils): opt-in chunked demean_map dispatch + corrected solver-phase memory attribution #605).
• JOSS paper finalization - 20 estimators, required sections (docs(paper): finalize JOSS submission - 20 estimators, required sections, draft-pdf CI #604).
• SyntheticControl ADH-2015 section 4 tail diagnostics (feat(synthetic_control): ADH-2015 §4 tail diagnostics (regression_weights + sparse_synthetic_control) #608).
• solve_ols marshalling slimming, ~15.3->7.8 GB Rust-side high-water (perf(linalg): slim solve_ols marshalling - 15.3->7.8 GB rust-side footprint, -21% firm-panel fits, outputs preserved #607).
• Assorted: SpilloverDiDResults diagnostic routing (feat(diagnostic_report): register SpilloverDiDResults routing #599), SyntheticControl placebo/LOO infeasible-reason split (feat(synthetic_control): split cv-infeasible from failed in placebo/LOO diagnostics #598), EfficientDiD survey-weighted Silverman bandwidth (feat(efficient_did): survey-weighted Silverman bandwidth in conditional Omega* #594), HAD continuous-path cluster-robust SE (feat(had): cluster-robust SE on the continuous paths (Phase 2a) #596), zero-column rank contract (fix: define zero-column rank detection contract #578), DiD predict support contract (docs: clarify DiD predict support contract #580), LP-DiD carousel (Add LP-DiD launch carousel (local-projections crossover narrative) #597), CI SHA-pinning + CODEOWNERS (ci: SHA-pin all GitHub Actions and add CODEOWNERS (supply-chain hardening) #595).

Methodology references (required if estimator / math changes)

• Method name(s): N/A - release chore. Only version strings and the CHANGELOG changed; no estimator or math logic is touched in this PR. (The underlying methodology changes being released already merged and were documented in their own PRs / docs/methodology/REGISTRY.md.)
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No test changes - version-string + CHANGELOG only.
• Backtest / simulation / notebook evidence (if applicable): N/A

Note on version choice

Cut as a patch (3.6.2) per maintainer direction. The FE-absorption bincount rewrite (#601) carries a documented ~1e-10 numerical-contract change and up-to-~1e-5 point-estimate shifts on absorbed 2x2 designs (see CHANGELOG "Changed"). Flagging for awareness - a minor bump would also be defensible.

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment: ✅ Looks good

Executive Summary

• Version metadata is consistently bumped to 3.6.2 across Python, packaging, Rust, citation, and guide surfaces: diff_diff/__init__.py:L304, pyproject.toml:L7, rust/Cargo.toml:L3, CITATION.cff:L10-L11, diff_diff/guides/llms-full.txt:L5.
• The changelog release header is set to ## [3.6.2] - 2026-07-03 and includes the comparison link: CHANGELOG.md:L8, CHANGELOG.md:L2083.
• No estimator implementation, weighting, variance/SE, inference, identification assumption, or default behavior is changed by this PR.
• Remaining 3.6.1 references are historical changelog/comparison text or benchmark baselines, not active package metadata.
• No P0/P1 findings.

Methodology
No findings.

Severity: N/A
Impact: The PR does not change causal estimators, formulas, assumptions, weighting, variance, inference, or defaults. Methodology cross-check against docs/methodology/REGISTRY.md is not triggered beyond confirming no methodology-bearing files are modified.
Concrete fix: None.

Code Quality
No findings.

Severity: N/A
Impact: Version-only edits are simple and consistent.
Concrete fix: None.

Performance
No findings.

Severity: N/A
Impact: No runtime code path changes.
Concrete fix: None.

Maintainability
No findings.

Severity: N/A
Impact: Version surfaces reviewed are synchronized.
Concrete fix: None.

Tech Debt
No findings.

Severity: N/A
Impact: No new deferred technical debt introduced.
Concrete fix: None.

Security
No findings.

Severity: N/A
Impact: No secrets, credential handling, network behavior, or security-sensitive code changed.
Concrete fix: None.

Documentation/Tests
No findings.

Severity: N/A
Impact: Changelog, citation metadata, and guide version are updated for the release. No tests were run; for this metadata-only bump, I do not consider that a blocker.
Concrete fix: None.