eval-harness

dataset

scifact_200

200 items · 26 arms. Headline: each arm's own primary metric, mean over its items; arms with different primary metrics are not ranked against each other.

baseline, promoted 2026-10-25

sf_e5_base_n200_v1

Build e5_base retrieval first: 0.7191 at $0 per query, not significantly different from any of the twelve BM25-then-rerank pipelines; add a reranker on its top 20 only if about +0.07 nDCG@10 is worth about $732 per 1M queries. REPORT_RETRIEVAL, executive summary.

Inspect the 26 arms on ndcg_10, scored against gold/scifact_200Open its sheet

goals: Evidence retrieval for scientific claims (SciFact)

01arms

scored against silverscored by a judgewrote the reference it is scored againstscored against a label set, not a reference setunmarked rows are scored against gold

26 shown
armmetricmeanitemscost per itemspeed per itemrunsscored against
sf_glm_s_e5first_n200_v1 · pairndcg_100.789200$0.000732 · 4,178 tokp50 3,007 · p95 4,028 ms1gold/scifact_200 goldinspect
sf_qwen_s_e5first_n200_v1 · pairndcg_100.771200$0.000754 · 4,485 tokp50 4,102 · p95 13,396 ms1gold/scifact_200 goldinspect
sf_glm_s_n200_v1ndcg_100.744200$0.000711 · 4,182 tokp50 3,004 · p95 3,980 ms1gold/scifact_200 goldinspect
sf_glm_s_bm25first_n200_v1 · pairndcg_100.742200$0.000741 · 4,182 tokp50 3,019 · p95 3,653 ms1gold/scifact_200 goldinspect
sf_glm_m_n200_v1ndcg_100.742200$0.002288 · 4,171 tokp50 7,430 · p95 10,170 ms1gold/scifact_200 goldinspect
sf_deepseek_m_n200_v1ndcg_100.740200$0.001190 · 3,869 tokp50 773 · p95 2,389 ms1gold/scifact_200 goldinspect
sf_deepseek_s_n200_v1ndcg_100.738200$0.000300 · 3,872 tokp50 2,392 · p95 7,323 ms1gold/scifact_200 goldinspect
sf_gemma_m_n200_v1ndcg_100.736200$0.000361 · 4,165 tokp50 4,730 · p95 13,747 ms1gold/scifact_200 goldinspect
sf_gemma_l_n200_v1ndcg_100.736200$0.000850 · 4,176 tokp50 8,641 · p95 17,014 ms1gold/scifact_200 goldinspect
sf_llama_l_n200_v1ndcg_100.728200$0.000865 · 3,977 tokp50 3,701 · p95 9,656 ms1gold/scifact_200 goldinspect
sf_e5_base_n200_v1 baselinendcg_100.719200$0.000000p50 56 · p95 75 ms1gold/scifact_200 goldinspect
sf_qwen_s_n200_v1 · repeatsndcg_100.719200$0.000779 · 4,514 tokp50 4,298 · p95 12,577 ms1gold/scifact_200 goldinspect
sf_qwen_s_n200_v1ndcg_100.718200$0.000779 · 4,514 tokp50 4,050 · p95 12,207 ms1gold/scifact_200 goldinspect
sf_llama_s_n200_v1ndcg_100.717200$0.000729 · 4,013 tokp50 13,661 · p95 42,002 ms1gold/scifact_200 goldinspect
sf_llama_l_n200_v1 · repeatsndcg_100.715200$0.000846 · 3,981 tokp50 10,149 · p95 28,200 ms1gold/scifact_200 goldinspect
sf_qwen_s_bm25first_n200_v1 · pairndcg_100.714200$0.000775 · 4,505 tokp50 4,032 · p95 12,558 ms1gold/scifact_200 goldinspect
sf_llama_s_n200_v1 · repeatsndcg_100.713200$0.000524 · 4,014 tokp50 3,903 · p95 10,564 ms1gold/scifact_200 goldinspect
sf_bge_small_n200_v1ndcg_100.710200$0.000000p50 26 · p95 45 ms1gold/scifact_200 goldinspect
sf_llama_m_n200_v1ndcg_100.709200$0.000717 · 4,109 tokp50 3,734 · p95 7,923 ms1gold/scifact_200 goldinspect
sf_mistral_s_n200_v1ndcg_100.707200$0.000426 · 4,191 tokp50 6,068 · p95 13,800 ms1gold/scifact_200 goldinspect
sf_gemma_s_n200_v1ndcg_100.702200$0.000481 · 4,119 tokp50 6,611 · p95 11,109 ms1gold/scifact_200 goldinspect
sf_mpnet_n200_v1ndcg_100.664200$0.000000p50 61 · p95 92 ms1gold/scifact_200 goldinspect
sf_minilm_n200_v1ndcg_100.653200$0.000000p50 16 · p95 23 ms1gold/scifact_200 goldinspect
sf_bm25_n200_v1ndcg_100.645200$0.000000p50 20 · p95 38 ms1gold/scifact_200 goldinspect
sf_random_n200_v1ndcg_100.001200$0.000000p50 1.6 · p95 1.9 ms1gold/scifact_200 goldinspect
sf_first_k_n200_v1ndcg_100.000200$0.000000p50 0.23 · p95 0.38 ms1gold/scifact_200 goldinspect

02what's inside

200 items, text, 4–28 words (median 12, middle half 9–15).

Source: BeIR/scifact (queries split). The text stays there; each item below links to its row.

025504–6 words: 147–9 words: 5010–12 words: 4513–15 words: 4616–18 words: 1419–21 words: 1822–24 words: 1025–27 words: 228–30 words: 14–67–910–1213–1516–1819–2122–2425–2728–30

the gold

gold/scifact_200: a list per item, 1–5 entries, median 1

200 shown
itemwordsgoldopen
10081 entriessource ↗ · inspect
101291 entriessource ↗ · inspect
101491 entriessource ↗ · inspect
1020221 entriessource ↗ · inspect
1021221 entriessource ↗ · inspect
1024111 entriessource ↗ · inspect
1029213 entriessource ↗ · inspect
1041152 entriessource ↗ · inspect
1049111 entriessource ↗ · inspect
1088121 entriessource ↗ · inspect
109941 entriessource ↗ · inspect
1104241 entriessource ↗ · inspect
1107101 entriessource ↗ · inspect
111081 entriessource ↗ · inspect
1121131 entriessource ↗ · inspect
113131 entriessource ↗ · inspect
1130171 entriessource ↗ · inspect
1144171 entriessource ↗ · inspect
115111 entriessource ↗ · inspect
1163101 entriessource ↗ · inspect
117581 entriessource ↗ · inspect
118121 entriessource ↗ · inspect
1185281 entriessource ↗ · inspect
1187231 entriessource ↗ · inspect
1191111 entriessource ↗ · inspect
1194211 entriessource ↗ · inspect
1196121 entriessource ↗ · inspect
1199181 entriessource ↗ · inspect
1200211 entriessource ↗ · inspect
1204141 entriessource ↗ · inspect
1207201 entriessource ↗ · inspect
1213131 entriessource ↗ · inspect
1216111 entriessource ↗ · inspect
122581 entriessource ↗ · inspect
124131 entriessource ↗ · inspect
1245101 entriessource ↗ · inspect
1259211 entriessource ↗ · inspect
1262121 entriessource ↗ · inspect
1266231 entriessource ↗ · inspect
127101 entriessource ↗ · inspect
1270141 entriessource ↗ · inspect
1271211 entriessource ↗ · inspect
1272121 entriessource ↗ · inspect
1273111 entriessource ↗ · inspect
1274253 entriessource ↗ · inspect
1278151 entriessource ↗ · inspect
1279121 entriessource ↗ · inspect
12881 entriessource ↗ · inspect
1280141 entriessource ↗ · inspect
1281101 entriessource ↗ · inspect
1282151 entriessource ↗ · inspect
129151 entriessource ↗ · inspect
1292101 entriessource ↗ · inspect
1298231 entriessource ↗ · inspect
13101 entriessource ↗ · inspect
130151 entriessource ↗ · inspect
130371 entriessource ↗ · inspect
1316101 entriessource ↗ · inspect
1319101 entriessource ↗ · inspect
13261 entriessource ↗ · inspect
1320161 entriessource ↗ · inspect
133195 entriessource ↗ · inspect
133591 entriessource ↗ · inspect
133681 entriessource ↗ · inspect
1339131 entriessource ↗ · inspect
1359201 entriessource ↗ · inspect
136281 entriessource ↗ · inspect
137131 entriessource ↗ · inspect
1379164 entriessource ↗ · inspect
138561 entriessource ↗ · inspect
138981 entriessource ↗ · inspect
1395211 entriessource ↗ · inspect
142201 entriessource ↗ · inspect
143171 entriessource ↗ · inspect
146181 entriessource ↗ · inspect
14851 entriessource ↗ · inspect
16391 entriessource ↗ · inspect
17974 entriessource ↗ · inspect
180191 entriessource ↗ · inspect
18591 entriessource ↗ · inspect
19851 entriessource ↗ · inspect
21271 entriessource ↗ · inspect
213141 entriessource ↗ · inspect
21691 entriessource ↗ · inspect
21881 entriessource ↗ · inspect
21981 entriessource ↗ · inspect
232121 entriessource ↗ · inspect
236101 entriessource ↗ · inspect
23981 entriessource ↗ · inspect
24871 entriessource ↗ · inspect
24971 entriessource ↗ · inspect
26951 entriessource ↗ · inspect
275143 entriessource ↗ · inspect
294121 entriessource ↗ · inspect
295181 entriessource ↗ · inspect
303131 entriessource ↗ · inspect
312141 entriessource ↗ · inspect
32451 entriessource ↗ · inspect
343152 entriessource ↗ · inspect
350141 entriessource ↗ · inspect
362241 entriessource ↗ · inspect
380121 entriessource ↗ · inspect
385132 entriessource ↗ · inspect
386161 entriessource ↗ · inspect
38891 entriessource ↗ · inspect
399111 entriessource ↗ · inspect
41091 entriessource ↗ · inspect
41191 entriessource ↗ · inspect
415131 entriessource ↗ · inspect
421131 entriessource ↗ · inspect
431121 entriessource ↗ · inspect
436131 entriessource ↗ · inspect
439131 entriessource ↗ · inspect
440131 entriessource ↗ · inspect
44391 entriessource ↗ · inspect
475101 entriessource ↗ · inspect
478241 entriessource ↗ · inspect
48141 entriessource ↗ · inspect
491141 entriessource ↗ · inspect
5071 entriessource ↗ · inspect
50171 entriessource ↗ · inspect
502161 entriessource ↗ · inspect
507151 entriessource ↗ · inspect
5191 entriessource ↗ · inspect
514191 entriessource ↗ · inspect
51781 entriessource ↗ · inspect
521251 entriessource ↗ · inspect
527171 entriessource ↗ · inspect
528241 entriessource ↗ · inspect
53101 entriessource ↗ · inspect
53271 entriessource ↗ · inspect
53371 entriessource ↗ · inspect
53591 entriessource ↗ · inspect
53961 entriessource ↗ · inspect
54082 entriessource ↗ · inspect
54491 entriessource ↗ · inspect
54971 entriessource ↗ · inspect
552201 entriessource ↗ · inspect
56151 entriessource ↗ · inspect
560131 entriessource ↗ · inspect
57151 entriessource ↗ · inspect
577241 entriessource ↗ · inspect
589211 entriessource ↗ · inspect
59773 entriessource ↗ · inspect
598211 entriessource ↗ · inspect
619152 entriessource ↗ · inspect
623131 entriessource ↗ · inspect
637131 entriessource ↗ · inspect
64192 entriessource ↗ · inspect
649131 entriessource ↗ · inspect
65971 entriessource ↗ · inspect
66061 entriessource ↗ · inspect
674111 entriessource ↗ · inspect
690191 entriessource ↗ · inspect
692101 entriessource ↗ · inspect
693101 entriessource ↗ · inspect
7062 entriessource ↗ · inspect
700111 entriessource ↗ · inspect
702121 entriessource ↗ · inspect
715151 entriessource ↗ · inspect
721131 entriessource ↗ · inspect
723151 entriessource ↗ · inspect
728122 entriessource ↗ · inspect
729111 entriessource ↗ · inspect
75151 entriessource ↗ · inspect
759191 entriessource ↗ · inspect
768111 entriessource ↗ · inspect
775161 entriessource ↗ · inspect
781141 entriessource ↗ · inspect
783131 entriessource ↗ · inspect
785151 entriessource ↗ · inspect
800191 entriessource ↗ · inspect
80571 entriessource ↗ · inspect
80891 entriessource ↗ · inspect
82181 entriessource ↗ · inspect
831121 entriessource ↗ · inspect
83261 entriessource ↗ · inspect
834101 entriessource ↗ · inspect
83781 entriessource ↗ · inspect
839141 entriessource ↗ · inspect
847181 entriessource ↗ · inspect
85971 entriessource ↗ · inspect
87041 entriessource ↗ · inspect
879101 entriessource ↗ · inspect
88081 entriessource ↗ · inspect
882101 entriessource ↗ · inspect
887121 entriessource ↗ · inspect
90391 entriessource ↗ · inspect
907171 entriessource ↗ · inspect
911141 entriessource ↗ · inspect
91361 entriessource ↗ · inspect
91471 entriessource ↗ · inspect
921101 entriessource ↗ · inspect
922121 entriessource ↗ · inspect
93671 entriessource ↗ · inspect
956131 entriessource ↗ · inspect
96051 entriessource ↗ · inspect
96792 entriessource ↗ · inspect
97591 entriessource ↗ · inspect
993101 entriessource ↗ · inspect

Each row is the profile evh dataset-items --dataset-id scifact_200 prints; inspect opens the item in the inspector with the dataset's largest group of comparable arms, and its text where the inspector runs on a machine that has it.

03reference sets

04labels

No labels yet. A sheet to fill: evh label-sheet --dataset-id scifact_200 --out sheet.csv; a run's outputs as a model set: evh label-from-run --run R --set S.