filipredpine commited on
Commit
cd5ba1c
·
verified ·
1 Parent(s): 1fb6ea5

Add the Redpine Biomedical QA question set

Browse files
Files changed (2) hide show
  1. README.md +86 -0
  2. data/test.jsonl +0 -0
README.md ADDED
@@ -0,0 +1,86 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-4.0
3
+ language:
4
+ - en
5
+ pretty_name: Redpine Biomedical QA
6
+ size_categories:
7
+ - n<1K
8
+ task_categories:
9
+ - question-answering
10
+ tags:
11
+ - biomedical
12
+ - medical
13
+ - literature
14
+ - retrieval-augmented-generation
15
+ - benchmark
16
+ configs:
17
+ - config_name: default
18
+ data_files:
19
+ - split: test
20
+ path: data/test.jsonl
21
+ ---
22
+
23
+ # Redpine Biomedical QA
24
+
25
+ 179 expert-validated biomedical questions, each written from one paper in the Redpine
26
+ Science full-text corpus and paired with a gold answer: two to four required claims, a
27
+ certainty sentence, the population and setting the claims hold for, and one over-claim
28
+ an answer must not make. It is the expert-validated question set of the preprint
29
+ [Leveraging a four-quadrant approach for evaluating Redpine Science](https://arxiv.org/abs/2610.07937).
30
+
31
+ The runner, the judges, the run log and the published verdicts are on GitHub:
32
+ [redpine-ai/benchmarks](https://github.com/redpine-ai/benchmarks/tree/main/science/redpine-biomedical-qa).
33
+
34
+ ```python
35
+ from datasets import load_dataset
36
+
37
+ ds = load_dataset("Redpine-AI/redpine-biomedical-qa", split="test")
38
+ ```
39
+
40
+ ## Fields
41
+
42
+ | Field | Meaning |
43
+ | --- | --- |
44
+ | `id` | stable id, `steered/` followed by the source paper's DOI |
45
+ | `track` | cardiology (79), neuroimmunology (60), sepsis (20) or immunology (20) |
46
+ | `question` | the question as reviewed |
47
+ | `required_claims` | the claims a strong answer conveys, in gold order (633 in all) |
48
+ | `certainty` | how certain the evidence is |
49
+ | `scope` | the population and setting the claims hold for |
50
+ | `wrong_if` | the over-claim an answer must not make; null for `steered/10.1177/15230864251401591` |
51
+ | `gold` | the full answer key as one string; the four fields above are parsed from it |
52
+ | `doi` | the source paper's DOI |
53
+ | `access` | `open` (105) or `paywalled` (74), the access class of the source paper |
54
+
55
+ Every `gold` reads:
56
+
57
+ A strong answer conveys: <claim>; <claim>[; <claim>[; <claim>]]. Certainty: <sentence>. Scope: <population and setting>. An answer is wrong if <over-claim>.
58
+
59
+ ## Scoring
60
+
61
+ The benchmark scores an answer per required claim: a fixed judge model marks each claim
62
+ `stated`, `contradicted` or `not_addressed`, and the answer gets claim coverage, a
63
+ contradiction rate and a correctness score. The judge prompt and the scoring rules are in
64
+ [judging.md](https://github.com/redpine-ai/benchmarks/blob/main/science/redpine-biomedical-qa/judging.md).
65
+
66
+ ## Provenance
67
+
68
+ Each question was written from one paper and reviewed by a domain expert; of 180 reviewed
69
+ questions, one was dropped. 125 questions were kept as written, 40 were reworded in the
70
+ expert's wording, and 14 had their gold edited.
71
+
72
+ Expert names, the experts' per-question review outcomes and comments, and the quoted
73
+ evidence passages behind each claim are not published; the DOI identifies the source paper.
74
+
75
+ ## Citation
76
+
77
+ ```bibtex
78
+ @misc{dorm2026redpine,
79
+ title = {Leveraging a four-quadrant approach for evaluating Redpine Science},
80
+ author = {Dorm, Filip and Vesterbacka, Leonora},
81
+ year = {2026},
82
+ eprint = {2610.07937},
83
+ archivePrefix = {arXiv},
84
+ url = {https://arxiv.org/abs/2610.07937}
85
+ }
86
+ ```
data/test.jsonl ADDED
The diff for this file is too large to render. See raw diff