Table 4.

IR Results on epic-qa-dev for Passage vs. Document retrieval experiments.

(a) IR Results on epic-qa-dev for document recall: Passages vs. Documents. 5000 passages ranking vs. 5000 document rankings.
Index	reranking	Test EPIC-QA_docs
		R@500	R@1K	R@2K	R@3K	R@4K	R@5K
passages	yes	0.6959	0.7979	0.8597	0.8693	0.8716	0.8716
documents	yes	0.7041	0.7867	0.8342	0.8538	0.8655	0.8686

(b) IR Results on epic-qa-dev for passage recall: Passages vs. Documents. 50,000 passages ranking vs. 5000 document ranking.
Index	Test EPIC-QA_passages
	R@100	R@500	R@1K	R@2K	R@5K	R@1K0	R@15K	R@30K	R@50K
pas	0.2199	0.4645	0.582	0.6815	0.7577	0.8321	0.8676	0.9027	0.9082
docs	0.0867	0.2234	0.3336	0.4345	0.6189	0.7442	0.7883	0.8422	0.8724

(c) IR Results on epic-qa-dev for nugget recall: Passages vs. Documents. 50,000 passages ranking vs. 5000 document ranking.
Index	Test EPIC-QA_passages
	N@100	N@500	N@1K	N@2K	N@5K	N@10K	N@15K	N@30K	N@50K
pas	0.5988	0.7934	0.88	0.9134	0.941	0.9589	0.9626	0.9774	0.9774
docs	0.3068	0.5156	0.6537	0.758	0.8661	0.921	0.942	0.9629	0.9644

(d) EPIC-QA evaluation results on epic-qa-dev: Passages vs. Documents. EPIC-QA metrics. 50,000 passages ranking vs. 5000 document ranking.
NDNS@k	Test EPIC-QA - epicQA evaluation
	pas			doc2pas			pas2sent
	Partial	Relaxed	Exact	Partial	Relaxed	Exact	Partial	Relaxed	Exact
1,000	0.2463	0.2296	0.1823	0.1124	0.1031	0.0936	0.1786	0.1799	0.2072
5,000	0.2519	0.2347	0.1871	0.1299	0.1180	0.1084	0.2098	0.2113	0.2425
15,000	0.2541	0.2362	0.1886	0.1338	0.1213	0.1118	0.2250	0.2266	0.2601
30,000	0.2552	0.2371	0.1893	0.1357	0.1228	0.1133	0.2274	0.2290	0.2628