Skip wall-clock timing tests on CI to stop false-positive flake

[igerber]

Summary

Follow-up to PR #330. That PR marked three wall-clock timing tests @pytest.mark.slow, which the default addopts = \"-m 'not slow'\" already excludes — so the default Python CI matrix became quiet. But the Rust-backend CI jobs at .github/workflows/rust-test.yml:155, 162, 190 explicitly override with -m '' to pull the full slow suite in (intentional — TROP parity tests need it). That's why PR #334's Windows py3.11 Rust job tripped a 0.120s vs 0.1s threshold failure that was pure runner noise.

This PR layers a skipif(os.environ.get(\"CI\") == \"true\", ...) marker on top of the existing @pytest.mark.slow. GitHub Actions sets CI=true on every runner, so the skip catches both invocation patterns. Local development (pytest, pytest -m slow, pytest -m '') is unaffected.

• test_se_accuracy.py::TestCallawaySantAnnaSEAccuracy::test_timing_performance
• test_se_accuracy.py::TestPerformanceRegression (class-level)
• test_methodology_honest_did.py::TestOptimalFLCI::test_m0_short_circuit

The test_m0_short_circuit case is the one PR #330 left untouched because it uses wall-clock as a correctness proxy for the short-circuit path (fast <0.5s, slow would blow past). The existing TODO.md entry already tracks replacing it with a mock/spy; skipif(CI) is the interim guard until that refactor lands.

Methodology references (required if estimator / math changes)

• Method name(s): N/A - no methodology changes
• Paper / source link(s): N/A
• Any intentional deviations from the source (and why): None

Validation

• Tests added/updated: No new tests; existing tests gain a second skip marker.
• Verified locally: `CI=true pytest … -m ''` reports 5 skipped (all three targets); `env -u CI pytest …` runs and passes the two cheap cases.
• 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

✅ Looks good — no unmitigated P0/P1 findings.

Executive Summary

• No methodology-impacting code changed. The diff only adds CI-only skip markers and explanatory comments in test files covering CallawaySantAnna and HonestDiD; it does not modify estimator code, assumptions, weighting, variance/SE, or defaults. Cross-check against docs/methodology/REGISTRY.md:L294-L456 and docs/methodology/REGISTRY.md:L2406-L2455 found no methodology defect.
• Pattern-consistency check passed: the repo’s actual wall-clock threshold assertions are the ones updated here, so this is not a partial fix.
• P3: the PR intentionally removes PR-time CI enforcement of CallawaySantAnna wall-clock budgets by skipping those tests whenever CI=="true" in tests/test_se_accuracy.py:L266-L286 and tests/test_se_accuracy.py:L422-L457.
• P3 (tracked): HonestDiD’s M=0 short-circuit check remains a timing-proxy test, now skipped on CI in tests/test_methodology_honest_did.py:L248-L275, but that limitation is already tracked in TODO.md:L110-L110 and non-timing M=0 correctness coverage still exists in tests/test_methodology_honest_did.py:L85-L100.

Methodology

• No findings. Cross-checked the affected method families against docs/methodology/REGISTRY.md:L294-L456 and docs/methodology/REGISTRY.md:L2406-L2455; this PR is test-only and does not introduce any undocumented deviation in estimator math, identification, weighting, or inference.

Code Quality

• No findings.

Performance

• Severity: P3. Impact: default pytest already excludes slow in pyproject.toml:L100-L107, but Rust CI explicitly re-includes the full suite with -m '' in .github/workflows/rust-test.yml:L142-L190. The new skipif(CI=="true") markers in tests/test_se_accuracy.py:L266-L286 and tests/test_se_accuracy.py:L422-L457 stop flaky failures, but they also mean CI no longer detects performance regressions for these wall-clock budgets. Concrete fix: keep the CI skip, but move these checks to a dedicated perf job or replace wall-clock assertions with a deterministic signal.

Maintainability

• No findings.

Tech Debt

• Severity: P3. Impact: tests/test_methodology_honest_did.py:L248-L275 still uses elapsed wall-clock time as a proxy for the M=0 short-circuit path and now skips that proxy on CI. This reduces automated runtime-path coverage, but it does not create a silent correctness bug, and the limitation is already tracked in TODO.md:L110-L110. Concrete fix: replace the timing proxy with a mock/spy on the optimizer call or an explicit short-circuit indicator, then remove the CI skip.

Security

• No findings.

Documentation/Tests

• No additional findings beyond the P3 items above. The new comments accurately match the current test-selection behavior documented in pyproject.toml:L100-L107 and the Rust CI workflow in .github/workflows/rust-test.yml:L142-L190.