pkg/agentdrain/anomaly_test.go used raw t.Error/t.Fatal calls and four separate single-scenario test functions, making failures hard to read and edge cases uncovered.

Changes

• Merged TestAnomalyDetection_NewTemplate/LowSimilarity/RareCluster/Normal into a single table-driven TestAnomalyDetector_Analyze with require/assert and descriptive messages
• Added 5 new sub-cases covering previously untested scenarios:
  • Similarity exactly at threshold (0.4 < 0.4 → false, boundary)
  • Similarity just below threshold (0.39 < 0.4 → true, boundary)
  • Combined LowSimilarity + RareCluster (score = 0.5)
  • nil cluster (guards the if cluster != nil branch explicitly)
  • Max achievable score: IsNewTemplate + RareCluster → 0.65 (with explanation that LowSimilarity is mutually exclusive with isNew=true)
• Converted TestAnalyzeEvent to require.NoError/require.NotNil/assert.True/assert.False with descriptive messages
• Added exact AnomalyScore assertions (assert.InDelta) instead of just > 0 checks, pinning the weighted scoring math

// Before – opaque failure message
if report.AnomalyScore <= 0 {
    t.Errorf("expected positive anomaly score for new template, got %v", report.AnomalyScore)
}

// After – precise, self-documenting
assert.InDelta(t, 0.65, report.AnomalyScore, 1e-9, "AnomalyScore mismatch")