CE Test Audit

You are auditing the test suite for quality issues. This skill applies the criteria from ADR-030, tests/README.md, and the test-quality-method docs. Load these references before auditing:

• references/adr-030-test-quality.md
• docs/improvement/test-quality-method/README.md
• docs/improvement/test-quality-method/anti_pattern_auditor.md

Focus option handling

• Option A (Test-Focused): run the full audit toolchain, including redundancy analysis.
• Option B (Code-Focused): run only hard-gate and anti-pattern checks needed to support source-focused remediation.
• Option C (Full Cycle): run Option A and then re-check hard gates after code-focused changes.

Audit Toolchain

Choose the command set based on focus option.

Option A / Option C (full test audit)

# 1. Anti-pattern scan (primary audit)
python scripts/anti-pattern-analysis/detect_test_anti_patterns.py

# 2. Private member usage scan (hard gate in CI)
python scripts/anti-pattern-analysis/scan_private_usage.py --check

# 3. Over-testing / redundant test density report
python scripts/over_testing/over_testing_report.py

# 4. Redundant test detection
python scripts/over_testing/detect_redundant_tests.py

# 5. Marker hygiene check
python scripts/quality/check_marker_hygiene.py --check

# 6. No test-helper export check (hard gate)
python scripts/quality/check_no_test_helper_exports.py

Option B (code-focused support audit)

# 1. Anti-pattern scan
python scripts/anti-pattern-analysis/detect_test_anti_patterns.py

# 2. Private member usage scan (hard gate in CI)
python scripts/anti-pattern-analysis/scan_private_usage.py --check

# 3. No test-helper export check (hard gate)
python scripts/quality/check_no_test_helper_exports.py

Results from (3) and (4) are also published as CI artifacts under ci-main.yml's advisory over-testing job.

Issue Classification (ADR-030 Priority Order)

Priority 1 — Determinism (highest)

Symptoms: flaky tests, CI failures without code changes. Detection: look for unpatched datetime.now(), random, network I/O, unseeded NumPy/Python RNG, or time.sleep. Fix: monkeypatch, tmp_path, np.random.seed(42), mock network calls.

Priority 2 — Private Member Usage

Detection: scan_private_usage.py --check output. Fix decision tree:

Is the private helper a stable domain concept?
├── YES → Make it public (rename, add docstring, export, update tests)
└── NO  → Delete the direct test; test through the public caller instead

Priority 3 — Assertion Strength

Symptoms: tests that only check assert result is not None or assert "key" in dict. Fix: Replace with semantic domain assertions:

# BEFORE
assert "predict" in result

# AFTER
assert result['low'] <= result['predict'] <= result['high']

Priority 4 — Layering / Test Scope

Symptoms: unit tests with heavy I/O, integration tests in tests/unit/, missing @pytest.mark.slow on long-running tests. Detection: check_marker_hygiene.py --check output; manual scope review. Fix: move to correct directory, add markers.

Priority 5 — Fixture Discipline

Symptoms: deeply chained fixtures (>3 levels), copying fixture code between test files, anonymous fixtures. Fix: consolidate in conftest.py; reduce fixture depth; name fixtures clearly.

Priority 6 — Semantic Redundancy (over-testing)

Detection: over_testing_report.py shows tests with zero unique lines; detect_redundant_tests.py shows identical coverage fingerprints. Rule (ADR-030): 0 tests with 0 unique lines are allowed unless:

1. It's a @pytest.mark.parametrize case with a meaningful variation.
2. It's a regression test for a tracked issue (@pytest.mark.issue). Fix: delete or merge into a parametrized test.

Fallback Chain Violations

Any test triggering a fallback warning (UserWarning) that does NOT use enable_fallbacks is a violation. Symptoms in CI:

• "Execution plugin error; legacy sequential fallback engaged"
• "Parallel failure; forced serial fallback engaged"
• "Cache backend fallback: using minimal in-package LRU/TTL implementation"

Fix: either fix the underlying condition (preferred) or explicitly mark the test with enable_fallbacks and add pytest.warns(UserWarning).

Reporting Template

When reporting audit findings, use this structure:

## Test Audit Report — <module or scope>

### P1 — Determinism Issues
- [ ] <test name>: <issue description> → <fix>

### P2 — Private Member Usage
- [ ] <test name>: <private symbol accessed> → <fix>

### P3 — Assertion Strength
- [ ] <test name>: assertion is execution-only → replace with <semantic assertion>

### P4 — Layering
- [ ] <test file>: scope mismatch → move to <target path>

### P5 — Fixture Discipline
- [ ] <fixture name>: depth <N> → simplify

### P6 — Redundancy
- [ ] <test name>: 0 unique lines / identical fingerprint to <other test> → delete or parametrize

### Hard Gate Violations (fix before PR merge)
- [ ] Private-member scan violations
- [ ] No-test-helper-export violations
- [ ] 90% coverage gate status

Out of Scope

• Writing new tests from scratch (see ce-test-author).
• Auditing visualization tests (ADR-023 exemption; these are excluded from coverage gate — check their # pragma: no cover placement instead).

Evaluation Checklist

• Selected focus-option command set run and output reviewed.
• Issues classified by ADR-030 priority.
• Hard gate violations flagged separately.
• Fallback chain violations identified.
• Recommendations are concrete (test name + fix action).