OpenClaw Evidence Tracker
OpenClaw Evidence Tracker 是一个面向科研工作流的 OpenClaw 插件，用来为 LLM 生成的科研 idea 追踪证据链、检索相关文献，并验证每条 claim 是否被已有研究支持。
项目概览
LLM 在生成跨学科科研 idea 时，经常会给出听起来合理但证据来源不清的假设。这个插件的目标是把 idea 拆成可验证的 claim，再用文献检索和语义验证建立证据链，最后生成带有颜色分级的 Markdown 报告。
| Item | Content |
|---|---|
| Plugin | OpenClaw Evidence Tracker |
| Scenario | LLM-generated research idea validation |
| Output | Evidence chain report with GREEN, YELLOW, and RED verdicts |
| Repository | View source on GitHub |
Core Workflow
- Parse a research PDF and extract paper content.
- Generate or receive a research idea, hypothesis, and verifiable claims.
- Search academic literature for each claim.
- Verify the relationship between the claim and cited evidence.
- Produce a final report with detailed evidence blocks and summary statistics.
Tool Capabilities
| Tool | Purpose |
|---|---|
parse_research_pdf |
Extract text from an uploaded research PDF |
search_literature |
Search academic sources for related papers |
run_evidence_chain |
Run the full PDF to idea to verification to report workflow |
verify_single_claim |
Verify one specific factual claim |
Verification Design
The plugin combines deterministic citation lookup with LLM semantic verification. Public academic sources such as CrossRef, Semantic Scholar, arXiv, and DBLP are used for metadata and abstracts. Optional semantic literature search can be enabled through a rag.ac API key.
The final verdicts are grouped into three levels:
| Verdict | Meaning |
|---|---|
| GREEN | The literature clearly supports the claim |
| YELLOW | The evidence is partial, uncertain, or the claim may overreach |
| RED | The claim is unsupported, contradicted, or mis-cited |
Security Notes
The public copy does not include real API keys, GitHub tokens, or local LLM service credentials. Runtime credentials should be provided through OpenClaw configuration or environment variables such as RAG_AC_API_KEY.
Source Code
The sanitized plugin source is published in the GitHub Pages repository: