test: add edge-case tests for all-NaN columns

tests/test_contextual.py

@@ -1,5 +1,6 @@
 import pytest
 import pandas as pd
+import numpy as np
 from tsauditor.anomaly.contextual import audit_contextual_anomalies
 from tsauditor.report.summary import WARNING
 
@@ -92,3 +93,24 @@ def test_local_spike_fails_global_zscore(clean_financial_df):
 
     spike_issues = [i for i in issues if i.code == "ANO003" and i.column == "Price"]
     assert len(spike_issues) >= 1
+
+
+def test_all_nan_column_skipped():
+    """Column that is entirely NaN is skipped gracefully."""
+    dates = pd.date_range("2026-01-01", periods=20, freq="D")
+    df = pd.DataFrame(
+        {"all_nan": [np.nan] * 20, "valid": range(20)},
+        index=dates,
+    )
+    issues = audit_contextual_anomalies(df, domain="finance")
+    nan_issues = [i for i in issues if i.column == "all_nan"]
+    assert len(nan_issues) == 0
+
+
+def test_single_row_df():
+    """Single row DataFrame: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"val": [1.0]}, index=dates)
+    issues = audit_contextual_anomalies(df, domain="finance")
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_correlation.py

@@ -133,3 +133,12 @@ def test_few_observations_skipped():
     t = _iid_target(n, 8)
     df = pd.DataFrame({"target": t, "leak": t.shift(-1)}, index=_idx(n))
     assert audit_correlation_leakage(df, target="target", min_obs=30) == []
+
+
+def test_single_row_df():
+    """Single row DataFrame with target: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"target": [1.0], "feat": [2.0]}, index=dates)
+    issues = audit_correlation_leakage(df, target="target", min_obs=1)
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_equivalence.py

@@ -153,3 +153,12 @@ def test_scattered_nans_use_pairwise_overlap():
     issues = audit_equivalence(df, target="target")
     assert "leak" in {i.column for i in issues}
     assert next(i for i in issues if i.column == "leak").evidence["n_obs"] < n
+
+
+def test_single_row_df():
+    """Single row DataFrame with target: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"target": [1.0], "feat": [2.0]}, index=dates)
+    issues = audit_equivalence(df, target="target", min_obs=1)
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_missing.py

@@ -86,6 +86,25 @@ def test_non_numeric_columns_ignored():
     assert len(issues) == 0
 
 
+def test_all_nan_column_reported_as_missing():
+    """All-NaN column is correctly reported (100% missing rate, clustering).
+
+    Unlike anomaly detectors which skip all-NaN columns, the missing profiler
+    rightfully flags them — 100% missing is a genuine data quality issue.
+    """
+    dates = pd.date_range("2026-01-01", periods=10, freq="D")
+    df = pd.DataFrame(
+        {"all_nan": [np.nan] * 10, "valid": range(10)},
+        index=dates,
+    )
+    issues = audit_missing(df, domain="finance")
+    nan_issues = [i for i in issues if i.column == "all_nan"]
+    # PRF006: high missing rate (100%), PRF002: clustered missing
+    codes = {i.code for i in nan_issues}
+    assert "PRF006" in codes
+    assert nan_issues[0].evidence["missing_percentage"] == 100.0
+
+
 def test_sensor_domain_lower_threshold(sensor_df):
     # Case 7 — Sensor domain with 3 consecutive NaNs -> PRF002 (lower threshold).
     df = sensor_df.copy()
@@ -100,3 +119,12 @@ def test_sensor_domain_lower_threshold(sensor_df):
     assert len(cluster_issues) == 1
     assert cluster_issues[0].evidence["cluster_threshold"] == 3
     assert cluster_issues[0].evidence["longest_consecutive_run"] == 3
+
+
+def test_single_row_df():
+    """Single row DataFrame: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"val": [1.0]}, index=dates)
+    issues = audit_missing(df, domain="finance")
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_point.py

@@ -53,3 +53,23 @@ def test_audit_point_anomalies_finance_threshold(base_date_index):
     # In finance (5.0), 4.5 is not an outlier
     issues = audit_point_anomalies(df, domain="finance")
     assert len(issues) == 0
+
+
+def test_audit_point_anomalies_all_nan_column_skipped(base_date_index):
+    """Column that is entirely NaN is skipped gracefully."""
+    df = pd.DataFrame(
+        {"all_nan": [np.nan] * 100, "valid": np.random.default_rng(42).normal(0, 1, 100)},
+        index=base_date_index,
+    )
+    issues = audit_point_anomalies(df)
+    nan_issues = [i for i in issues if i.column == "all_nan"]
+    assert len(nan_issues) == 0
+
+
+def test_single_row_df():
+    """Single row DataFrame: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"val": [1.0]}, index=dates)
+    issues = audit_point_anomalies(df)
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_profiler.py

@@ -130,3 +130,12 @@ def test_sensor_domain_median_threshold(sensor_df):
     issue = gap_issues[0]
     assert issue.severity == WARNING
     assert issue.evidence["maximum_gap_days"] >= 1.0
+
+
+def test_single_row_df_missing():
+    """Single row DataFrame for frequency: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"val": [1.0]}, index=dates)
+    issues = audit_frequency(df, domain="finance")
+    assert isinstance(issues, list)
+    assert len(issues) == 0

tests/test_stationarity.py

@@ -76,3 +76,24 @@ def test_audit_stationarity_with_nan_and_inf(base_date_index):
 
     issues = audit_stationarity(df, min_obs=25)
     assert isinstance(issues, list)
+
+
+def test_all_nan_column_skipped(base_date_index):
+    """Column that is entirely NaN is skipped gracefully."""
+    df = pd.DataFrame(
+        {"all_nan": [np.nan] * 100, "valid": np.random.randn(100)},
+        index=base_date_index,
+    )
+    issues = audit_stationarity(df, min_obs=25)
+    nan_issues = [i for i in issues if i.column == "all_nan"]
+    assert len(nan_issues) == 0
+
+
+def test_single_row_df_raises_without_guard():
+    """Single row DataFrame passes min_obs=0 but statsmodels adfuller
+    rejects constant input — guard is missing (known issue)."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"val": [1.0]}, index=dates)
+    # TODO: remove pytest.raises once a min_obs guard is added upstream
+    with pytest.raises(ValueError, match="constant"):
+        audit_stationarity(df, min_obs=0)

tests/test_temporal.py

@@ -112,3 +112,12 @@ def test_few_observations_skipped():
     t = _ar1(n, seed=8)
     df = pd.DataFrame({"target": t, "leak": t.shift(-1)}, index=_idx(n))
     assert audit_temporal_leakage(df, target="target", min_obs=30) == []
+
+
+def test_single_row_df():
+    """Single row DataFrame with target: returns empty list, no crash."""
+    dates = pd.date_range("2026-01-01", periods=1, freq="D")
+    df = pd.DataFrame({"target": [1.0], "feat": [2.0]}, index=dates)
+    issues = audit_temporal_leakage(df, target="target", min_obs=1)
+    assert isinstance(issues, list)
+    assert len(issues) == 0