hypothesis-queue/examples/desktop-queue.json at main · lemonlavender/hypothesis-queue · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
{
  "schemaVersion": 1,
  "title": "AI-assisted documentation study / 文档研究",
  "budgetHours": 4,
  "hypotheses": [
    {
      "id": "H1",
      "claim": "Source-linked AI drafts reduce unsupported statements.",
      "source": "Synthetic pilot: 4 unsupported statements among 20 draft claims.",
      "prediction": "A source-linked prompt produces at most 1 unsupported claim in 20.",
      "falsifier": "More than 1 of 20 claims lacks a supporting passage.",
      "method": "Compare two prompts on the same 20 synthetic questions; a reviewer checks every cited passage.",
      "owner": "Research team A",
      "hours": 2,
      "priority": 1,
      "parked": false,
      "result": null
    },
    {
      "id": "H2",
      "claim": "A checklist reduces review time without reducing accuracy.",
      "source": "Synthetic review log: median 12 minutes per document.",
      "prediction": "Median time falls below 9 minutes while unsupported-claim recall stays at least 90%.",
      "falsifier": "Median time is 9 minutes or more, or recall is below 90%.",
      "method": "Counterbalance 10 paired documents and record time and missed claims.",
      "owner": "Research team B",
      "hours": 3,
      "priority": 1,
      "parked": false,
      "result": null
    },
    {
      "id": "H3",
      "claim": "Showing uncertainty improves reader decisions.",
      "source": "AI-generated proposal; no pilot evidence yet.",
      "prediction": "Readers distinguish unsupported statements more accurately.",
      "falsifier": "",
      "method": "",
      "owner": "",
      "hours": 1,
      "priority": 2,
      "parked": false,
      "result": null
    },
    {
      "id": "H4",
      "claim": "A small glossary reduces terminology mismatches.",
      "source": "Synthetic baseline: 5 mismatches across 30 term uses.",
      "prediction": "At most 2 mismatches across the same 30 uses.",
      "falsifier": "3 or more mismatches remain.",
      "method": "Blindly label each term use using the same glossary and count mismatches.",
      "owner": "Research team A",
      "hours": 1.5,
      "priority": 2,
      "parked": false,
      "result": null
    }
  ]
}