Human Rights Case Pathfinder

01

The user journey

One research chain, from story to reviewable pack.

• Describe events, dates, actors and current stage.
• Receive a factual issue map that separates the user’s words from possible legal characterization.
• See missing dates, documents, protected-ground facts and forum questions.
• Retrieve exact citations, pinpoints and passages from an authorized corpus.
• Compare material similarities and differences, including facts that weaken the match.
• Export a chronology, authority list, document checklist and questions for a lawyer or institution.

02

First public build

The foundation can work without cloud LLM fees.

A guided form collects events, dates, ground, forum and stage. An authorized local corpus is parsed into metadata and indexed with SQLite FTS5/BM25. Verified templates assemble the procedural roadmap. Each result stores its source, corpus version, query and date.

This proves the corpus, citations, export and evaluation set before generative inference adds cost and new failure modes. BC’s Civil Resolution Tribunal Solution Explorer is a useful guided, rules-based model.

SourcesCivil Resolution Tribunal

03

Funded AI layer

Meaningful multilingual access at scale does require inference capacity.

• Correctable natural-language fact extraction.
• English, French and Traditional Chinese intake tested for legal meaning.
• Semantic retrieval beside lexical search, with benchmarked recall.
• Decision comparison and source-linked plain-language summaries.
• Detection of missing, conflicting and adverse authority.
• Secure server-side inference, privacy controls, monitoring and independent evaluation.

SourcesYale News · Stanford Legal Design Lab · Stanford Legal Design Lab

04

Release gates

No feature ships because the demonstration looks fluent.

• Corpus authorization and a documented update route.
• Citation-validity tests against full primary sources.
• Retrieval tests that count relevant decisions missed.
• Procedural rules checked against the current official version.
• Privacy and security review for every data flow.
• Task testing with self-represented users and legal-help reviewers.
• Correction and rollback for sources, rules and model behaviour.

SourcesStanford Law School · Stanford Legal Design Lab

05

Funding proposition

The ask is an accountable service, not a bucket of tokens.

The Lab will approach partners with a published journey, source ledger, lawful-corpus plan, retrieval prototype and evaluation results. Funding would support authorized data, multilingual access, secure inference, independent testing, community participation and maintenance.

That is the basis for a useful conversation with Yale, Stanford, clinics and public institutions: a concrete workflow and evidence record they can challenge, improve or help distribute.

SourcesYale News · Stanford Legal Design Lab