implementation	model	family	comparison	n_unique_targets	mean_paired_difference	bootstrap_ci95_low	bootstrap_ci95_high	alternative	paired_sign_flip_p	bootstrap_replicates	sign_flip_permutations	holm_p_within_model_family
first_unbalanced_run	safe_external_v1	directional	raw_cosine_vs_train_mean	2160	0.001006855543252	-0.000349451675815	0.0023256907720466	greater	0.0688186236275274	10000	50000	0.0688186236275274
balanced_final_run	safe_external_v1	directional	raw_cosine_vs_train_mean	2160	0.0016365594825738	0.0003947280861909	0.0028060510205708	greater	0.0035599288014239	10000	50000	0.0035599288014239
