DIKWP ProofLedger OS

English documentation · 中文速览 · Download the release package

Release layout / 发布结构 — The runnable source is distributed in dikwp_proofledger_os_open_source_app.zip. Download and extract it first, enter the extracted application directory, and then run Quick start. / 可运行源码位于上述 ZIP 发布包中；请先下载并解压，进入应用目录，再执行下方命令。

中文速览

• 它解决什么问题：把 AI 回答、RAG 输出、报告或政策文本拆分为可检查的“主张—证据”账本，标记证据缺口、来源关系与待复核项。
• 适合谁：RAG 与大模型团队、研究者、审计与治理人员，以及需要对高风险内容进行人工复核的组织。
• 如何开始：解压发布包后运行示例验证命令；系统会生成主张表、证据映射、风险项和可复核输出。
• 使用边界：本项目不自动浏览网页、不保证真值，也不替代法律、科学或专业判断；它提供的是白盒核验脚手架与可逆审查流程。

DIKWP ProofLedger OS is an open-source claim-evidence verification layer for AI answers, RAG systems, reports, policies, brand content, and agent outputs.

It converts an AI answer into a DIKWP proof ledger:

• D / Data: raw answer text, source snippets, citations, extracted claims.
• I / Information: citation links, source-claim relations, overlap, conflict, missing evidence.
• K / Knowledge: claim type, proof status, source coverage, contradiction signals.
• W / Wisdom: risk, unsupported-strength flags, citation abuse, action boundary.
• P / Purpose: local purpose of the answer and the verification task.
• R / Reliability: Solid / Supported / Provisional / Borrowed / Hollow / Contested / Blocked.

The goal is not to make AI sound more confident. The goal is to make AI outputs auditable, demotable, correctable, and reversible.

Core use cases

• RAG citation integrity checks
• AI hallucination triage
• Claim-to-source alignment reports
• Enterprise AI answer governance
• Academic and policy report evidence ledgers
• Brand claim verification before AI SEO / GEO publishing
• Agent output review before tool execution or external publication

Quick start

pip install -e .
dikwp-proofledger verify examples/sample_ai_answer.md examples/sample_sources.json --out outputs/demo

Run static boundary audit:

dikwp-proofledger static-audit src --out outputs/demo/static_boundary_audit_report.json

Optional local dashboard:

pip install -e .[app]
streamlit run src/dikwp_proofledger/app.py

Outputs

A demo run generates:

• proofledger_report.json
• claim_cards.json
• claim_cards.csv
• evidence_alignment_matrix.csv
• source_coverage.json
• recommendations.md
• static_boundary_audit_report.json

What this app does not do

It does not browse the web, scrape sites, execute external tools, guarantee truth, or replace human/legal/scientific review. It is a local white-box verification scaffold.