Add workflow improvements to reduce PR review rounds

[igerber]

Summary

• Add safe_inference() utility to diff_diff/utils.py for NaN-safe inference computation (t_stat, p_value, CI computed together with NaN gate)
• Add assert_nan_inference() test helper to tests/conftest.py for validating NaN consistency across all inference fields
• Add 10 tests for safe_inference() covering NaN/zero/negative/Inf SE, valid SE with normal and t-distribution, return types
• Add prescriptive inference rules to CLAUDE.md (design pattern Prepare v0.2.0 release #7, test guidelines, checklist items)
• Enhance pre-merge-check patterns with 4 concrete grep-based checks (inline inference, zero-SE fallback, missing get_params, unguarded CI)
• Add methodology cross-check to plan review (Registry alignment in Dimension 4)
• Add known anti-patterns section to Codex review prompt (inline inference P1, missing downstream params P1, partial NaN guard P0)
• Integrate pre-commit pattern checks into submit-pr and push-pr-update skills
• Document safe_inference() migration of existing call sites as follow-up in TODO.md

Methodology references (required if estimator / math changes)

• Method name(s): N/A - safe_inference() is a utility wrapper around existing compute_p_value() and compute_confidence_interval(), no new methodology
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: tests/test_utils.py (10 new tests in TestSafeInference class), tests/conftest.py (new assert_nan_inference helper)
• Backtest / simulation / notebook evidence (if applicable): N/A

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall Assessment: ⚠️ Needs changes

Executive Summary

• Added safe_inference() with NaN gating plus tests; no estimator math/SE defaults changed.
• Added NaN inference test helper and workflow guidance updates.
• New workflow grep patterns use PCRE tokens without grep -P, so the checks likely won’t match and can miss inline inference/zero‑SE cases.
• Tests not run.

Methodology
No findings. Registry reviewed; this PR does not change estimator equations/assumptions or SE definitions, only adds the helper in diff_diff/utils.py:L150-L181.

Code Quality

1. Severity: P2. Impact: Pattern checks intended to catch inline inference and zero‑SE fallbacks will not match because \s/\b aren’t recognized by default grep, so the checks can silently miss methodology‑critical issues. Concrete fix: Use grep -P (PCRE) with the current regexes or switch to POSIX classes, e.g. grep -En "t_stat[[:space:]]*=[[:space:]]*[^#]*/[[:space:]]*se" and grep -En "if.*(se|SE).*>.*0.*else[[:space:]]+(0\\.0|0)" (and drop \\b). Update in .claude/commands/pre-merge-check.md:L34-L43, .claude/commands/push-pr-update.md:L124-L132, .claude/commands/submit-pr.md:L136-L144, TODO.md:L76-L79.

Performance
No findings.

Maintainability
No findings.

Tech Debt
No findings.

Security
No findings.

Documentation/Tests

1. Severity: P3. Impact: Changes rely on new tests but they were not run as part of this review, so behavior is unverified. Concrete fix: Run pytest tests/test_utils.py -k safe_inference or the full test suite.

[igerber]

/ai-review

[github-actions]

🔁 AI review rerun (requested by @igerber)

Head SHA: 5cde1b7e5d693d1308fc7c3e25db67fde513a549

---

Overall assessment: ✅ Looks good

Executive Summary

• Added safe_inference with NaN-gated inference outputs to prevent misleading stats when se is invalid. diff_diff/utils.py:L150-L181
• Tests cover NaN/zero/inf SEs, distribution choice, and CI behavior for safe_inference. tests/test_utils.py:L527-L621
• Added assert_nan_inference helper to standardize NaN-consistency checks across inference fields. tests/conftest.py:L141-L175
• No estimator/SE/identification or default behavior changes in estimator code paths.

Methodology
No findings. Reviewed docs/methodology/REGISTRY.md; no estimator math/SE/identification or default changes in this PR, and safe_inference is a new utility not wired into estimator paths.

Code Quality
No findings.

Performance
No findings.

Maintainability
No findings.

Tech Debt
No findings.

Security
No findings.

Documentation/Tests
No findings.

Tests not run (not requested).