func TestRetrievalRecallGate(t *testing.T)

		t.Fatalf("REGRESSION: mean recall@10 = %.3f is below the 0.8 floor; the hybrid search is no longer recalling the relevant candidates", meanRecall10)
	}
	// meanMRR >= 0.5: a healthy search puts a relevant candidate in the top two on average (rank one is 1.0,
	// rank two is 0.5), so 0.5 is a conservative floor on first relevant rank that a ranking regression breaks.
	if meanMRR < 0.5 {
		t.Fatalf("REGRESSION: mean MRR = %.3f is below the 0.5 floor; the first relevant candidate is ranking too deep", meanMRR)
	}
	// meanNDCG@10 >= 0.6: nDCG rewards ranking the relevant ids high, not merely present. 0.6 is a conservative
	// floor that a correct ordering clears while a fusion or ranking regression that buries relevant ids breaks.
	if meanNDCG10 < 0.6 {
		t.Fatalf("REGRESSION: mean nDCG@10 = %.3f is below the 0.6 floor; relevant candidates are ranking too low in the fused order", meanNDCG10)
	}
	// recall@5 is logged as a baseline and sanity checked but not floored as tightly, since a five deep cutoff
	// over a twelve candidate pool is the harder bar; it must still recall a clear majority on average.
	if meanRecall5 < 0.6 {
		t.Fatalf("REGRESSION: mean recall@5 = %.3f is below the 0.6 floor; the top five no longer surfaces the relevant candidates", meanRecall5)
	}

meanNDCG@10 >= 0.6: nDCG rewards ranking the relevant ids high, not merely present. 0.6 is a conservative