Release v2.7.2

[igerber]

Summary

• Bump version to 2.7.2 across __init__.py, pyproject.toml, and rust/Cargo.toml
• Populate CHANGELOG with all changes since v2.7.1:
  • Added: SEO infrastructure, AI discoverability (llms.txt/llms-full.txt)
  • Changed: TROP method="twostep" renamed to method="local", internal _joint_* → _global_*
  • Fixed: TROPResults schema (unit counts not observation counts), llms-full.txt accuracy

Methodology references (required if estimator / math changes)

• Method name(s): N/A — version bump only, no methodology changes in this PR
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No test changes (version bump only)
• All methodology changes were validated in their respective PRs (Rename TROP method="twostep" to method="local" #209, Add SEO infrastructure and AI discoverability #210)

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment
✅ Looks good

Highest unmitigated finding is P2. This PR is release metadata only, so there are no P0/P1 methodology or inference defects in scope.

Executive Summary

• P2: release-version propagation is incomplete. CITATION.cff:8 and docs/llms-full.txt:5 still report 2.7.1, while the package versions were bumped to 2.7.2 in diff_diff/init.py:182, pyproject.toml:7, and rust/Cargo.toml:3.
• No methodology-affecting code changed. After checking docs/methodology/REGISTRY.md, this diff does not touch estimators, weights, variance/SE, identification checks, or defaults.
• The edge-case checklist is not implicated here: no new data path, inference path, control-group logic, or inline inference code was introduced by the changed files.
• The release notes were added at CHANGELOG.md:10, but the llms-full.txt accuracy fix claimed at CHANGELOG.md:21 is not fully reflected because docs/llms-full.txt:5 is still stale.
• No performance, security, or runtime-code-quality concerns were introduced by the version bumps themselves.
• No test changes are needed for a metadata-only release PR.

Methodology

• No findings. I checked docs/methodology/REGISTRY.md; this PR only changes release notes and version literals in CHANGELOG.md:10, diff_diff/init.py:182, pyproject.toml:7, and rust/Cargo.toml:3.

Code Quality

• No findings. No executable logic changed, so the known anti-patterns around inline inference, partial NaN guards, control groups, and downstream parameter propagation are not introduced here.

Performance

• No findings. No runtime path or algorithm changed.

Maintainability

• No separate findings beyond the version-drift issue in Documentation/Tests.

Tech Debt

• No findings. Nothing in this diff needs TODO.md mitigation.

Security

• No findings. No secrets or security-sensitive code paths changed.

Documentation/Tests

• Severity: P2. Impact: release metadata is internally inconsistent. The package is bumped to 2.7.2 in diff_diff/init.py:182, pyproject.toml:7, and rust/Cargo.toml:3, but CITATION.cff:8 and docs/llms-full.txt:5 still advertise 2.7.1. That will publish stale citation metadata and stale AI-facing docs for the 2.7.2 release. Concrete fix: update those two files to 2.7.2 in this PR, or generate them from the canonical package version during release automation.
• No separate test finding. This PR does not alter executable behavior.