Shorten test suite runtime with parallel execution and reduced iterations

[igerber]

Summary

• Add pytest-xdist for parallel test execution (-n auto --dist worksteal) across all CI jobs
• Cap min_n at 49 in pure Python mode to reduce bootstrap iterations from 199-249 to 49
• Widen convergence tolerances conditionally when bootstrap iterations are reduced
• Refactor TestTROPResults to use class-scoped shared fixture, eliminating 7 redundant TROP fits
• Combine test_is_significant and test_significance_stars into single test
• Reduce simulation counts in test_power.py (410 → 190 total simulations)
• Reduce TROP methodology test data sizes (~60% fewer observations in LOOCV-heavy tests)
• Remove -x fail-fast flag from CI (incompatible with xdist parallel workers)

Methodology references (required if estimator / math changes)

• N/A - no methodology changes. Only test infrastructure and CI configuration changes.

Validation

• Tests added/updated: test_ci_params.py (2 updated, 2 new), test_staggered.py, test_methodology_callaway.py, test_trop.py, test_power.py
• Full test suite verified: 1035 passed, 30 skipped, 0 failures with xdist parallelization
• All convergence tests pass with wider tolerances under reduced bootstrap iterations

Security / privacy

• Confirm no secrets/PII in this PR: Yes

---

Generated with Claude Code

[github-actions]

Overall assessment: ✅ Looks good

Executive summary

• No estimator/methodology changes detected; registry entries and estimator docstrings for CallawaySantAnna and TROP remain unchanged. (docs/methodology/REGISTRY.md:L259-L360, docs/methodology/REGISTRY.md:L546-L680, diff_diff/staggered.py:L149-L219, diff_diff/trop.py:L345-L424)
• CI/test runtime reductions are confined to the test harness (bootstrap cap in pure-Python mode and reduced simulations). (tests/conftest.py:L107-L118, tests/test_power.py:L393-L533)
• Minor documentation mismatch: CLAUDE guidance prescribes a 0.15 SE‑convergence tolerance for n_boot >= 100, but CallawaySantAnna SE convergence tests use 0.25/0.20. (CLAUDE.md:L381-L381, tests/test_methodology_callaway.py:L803-L842, tests/test_staggered.py:L1582-L1590, tests/test_staggered.py:L1730-L1741)

Methodology

• No issues found. Changes are test/CI-only; estimator docstrings and registry requirements remain unaffected. (docs/methodology/REGISTRY.md:L259-L360, docs/methodology/REGISTRY.md:L546-L680, diff_diff/staggered.py:L149-L219, diff_diff/trop.py:L345-L424)

Code Quality

• No issues found.

Performance

• No issues found (production runtime unaffected).

Maintainability

• No issues found.

Tech Debt

• No issues found.

Security

• No issues found.

Documentation/Tests

• Severity: P3 — Impact: Test-writing guidance now specifies threshold = 0.40 if n_boot < 100 else 0.15, but CallawaySantAnna SE convergence tests use looser 0.25/0.20 thresholds. This inconsistency can mislead contributors and cause future updates to drift. — Concrete fix: Update the guidance to reflect the per-test tolerances (or tighten the tests to 0.15/0.20 if that’s the intended standard). (CLAUDE.md:L381-L381, tests/test_methodology_callaway.py:L803-L842, tests/test_staggered.py:L1582-L1590, tests/test_staggered.py:L1730-L1741)