verify_quotes.sh
4.82 KB
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
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
#!/usr/bin/env bash
# 引文回查脚本（C1 / benchmark_v2 附录 B R-G1~R-G4 的留痕工具）
# 用法: verify_quotes.sh → 对每条拟用引文做 grep -F 命中计数，输出 FILE<TAB>HITS<TAB>QUOTE
# 口径: HTML/MD 直读逐字（口径1）。空白不敏感命中（口径2）与 PDF 解码转写（口径3）见 verify_quotes.py
RAW=/app/system/agents/deep-research/train/hunt/sources_raw
r() { local f="$1"; local s="$2"; local n
n=$(grep -o -F -- "$s" "$RAW/$f" 2>/dev/null | wc -l)
printf '%s\t%s\t%s\n' "$f" "$n" "$s"; }
echo "### S1 anthropic-multiagent.html"
r anthropic-multiagent.html '15× more tokens than chats'
r anthropic-multiagent.html '90.2% on our internal research eval'
r anthropic-multiagent.html 'Simple fact-finding requires just 1 agent with 3-10 tool calls'
r anthropic-multiagent.html 'more than 10 subagents with clearly divided responsibilities'
r anthropic-multiagent.html 'start with short, broad queries'
r anthropic-multiagent.html "Each subagent needs an objective, an output format"
r anthropic-multiagent.html 'agents continuing when they already had sufficient results'
r anthropic-multiagent.html 'scouring the web endlessly for nonexistent sources'
r anthropic-multiagent.html 'Adding source quality heuristics'
r anthropic-multiagent.html "game of telephone"
r anthropic-multiagent.html 'saving its plan to Memory'
r anthropic-multiagent.html '3-5 subagents in parallel rather than serially'
r anthropic-multiagent.html 'use 3+ tools in parallel'
r anthropic-multiagent.html 'spawning 50 subagents for simple queries'
r anthropic-multiagent.html 'CitationAgent processes the documents'
r anthropic-multiagent.html 'do the cited sources match the claims'
r anthropic-multiagent.html 'did it use primary sources over lower-quality secondary sources'
r anthropic-multiagent.html 'Research demands the flexibility to pivot'
r anthropic-multiagent.html 'chose SEO-optimized content farms over authoritative'
r anthropic-multiagent.html 'examine all available tools first'
r anthropic-multiagent.html 'duplicated work investigating current 2025 supply chains'
r anthropic-multiagent.html 'match tool usage to user intent'
r anthropic-multiagent.html 'Research Lead'
r anthropic-multiagent.html 'citation accuracy'
echo "### S7/S8/S9/S10/S11/S17"
r gpt-researcher-readme.md "utilize 'planner' and 'execution'"
r gpt-researcher-readme.md 'publisher'
r gpt-researcher-readme.md 'the more sites we scrape the less chances of incorrect data'
r storm-readme.md 'generating long articles with citations into two steps'
r storm-readme.md 'Perspective-Guided Question Asking'
r storm-readme.md 'pre-writing'
r storm-readme.md 'Stanford OVAL'
r storm-paper-arxiv.html 'source bias transfer'
r storm-paper-arxiv.html 'over-association of unrelated facts'
r storm__full-paper.html 'source bias transfer'
r storm__full-paper.html 'over-association of unrelated facts'
r storm__full-paper.html 'citation recall'
r storm__full-paper.html 'citation precision'
r storm__full-paper.html 'we calculate the'
r storm__full-paper.html 'based on the definition in Gao et al'
r storm__full-paper.html '84.83'
r storm__full-paper.html '85.18'
r storm__full-paper.html 'Severely lacking'
r storm__full-paper.html 'Exemplary in breadth'
r mbfc-methodology.html 'a minimum of 10 headlines and 5 full news stories'
r mbfc-methodology.html 'no universally accepted scientific formula exists'
r mbfc-methodology.html 'remains primarily tailored to the political landscape of the United States'
r mbfc-methodology.html 'structured human assessments'
echo "### S14/S15/S16/S18/S19"
r openai__deep-research-mcp-server-README.md 'relies specifically on Search and Fetch tools'
r openai__deep-research-mcp-server-README.md "do not call fetch again for that same file"
r openai__deep-research-mcp-server-README.md 'Include inline citations and return all source metadata'
r openai__deep-research-mcp-server-README.md 'Prioritize reliable, up-to-date sources'
r google__gemini-deep-research.md 'creates a multi-step research plan for you to either revise or approve'
r google__gemini-deep-research.md 'comprehensive report with key findings and links to original sources'
r google__gemini-deep-research.md 'hundreds'
r google__gemini-deep-research.md 'browses'
r hapgood__SIFT-four-moves.md 'Stop, Investigate the source, find better coverage'
r hapgood__SIFT-four-moves.md 'Trace claims, quotes, and media back to the original context'
r hapgood__SIFT-four-moves.md 'stopping periodically and reevaluating our reaction or search strategy is key'
r hapgood__SIFT-four-moves.md 'Just as when you hear a rumor you want to know who the source is before reacting'
r open_deep_research__README.md '$45.98'
r open_deep_research__README.md '$187.09'
r open_deep_research__README.md 'RACE'
r open_deep_research__README.md 'Summarization'
r open_deep_research__README.md 'Compression'
r open_deep_research__README.md 'Final Report'