Table 2.

  Robustness to retrieval configuration (Mean ± Std Across Runs).

Dataset	Retrieval configuration	Accuracy degradation (%)
TruthfulQA	Baseline (k = 5)	21.86 ± 0.19
	No Filtering	20.92 ± 0.17
	k = 10	23.20 ± 0.14
	k = 3	21.10 ± 0.14
MMLU	Baseline (k = 5)	18.70 ± 0.14
	No Filtering	18.10 ± 0.14
	k = 10	19.90 ± 0.14
	k = 3	18.50 ± 0.14
MedMCQA	Baseline (k = 5)	23.90 ± 0.14
	No Filtering	22.80 ± 0.14
	k = 10	24.30 ± 0.14
	k = 3	23.10 ± 0.14
SCALR	Baseline (k = 5)	20.40 ± 0.14
	No Filtering	19.60 ± 0.14
	k = 10	21.70 ± 0.14
	k = 3	20.20 ± 0.14