Trust / Methodology

Citations.law is built around a simple professional rule: No source, no legal assertion.

The governing model

AI can suggest. Sources must support. Deterministic systems must verify. Attorneys must decide.

Our goal is not to pretend that software replaces legal judgment. The goal is to make legal work more inspectable before it reaches a judge, client, adversary, or file.

✓

What is deterministic

Citation parsing, citation normalization, canonical corpus lookup, source hashes, document hashes, report hashes, corpus-version labels, and audit metadata are handled as deterministic verification steps where possible.

What uses AI

AI may help extract document structure, identify candidate issues, summarize retrieved source law, propose drafts, and explain flagged problems. AI is assistive; it is not the final authority.

⚖

Attorney-controlled output

Reports label uncertainty and attorney-review items instead of hiding them. The lawyer remains responsible for legal judgment, filing decisions, and professional obligations.

Known limitations

Citations.law should underpromise with visible rigor. “Not found in current corpus” does not prove a citation is fake. Treatment signals and proposition-support review require careful lawyer review. Coverage depends on the configured corpus and disclosed sources.

Current verification is strongest for case reporter citation existence and identity.
Quote, pin-cite, treatment, adverse-authority, and court-rule checks are being built as separate services with explicit labels.
No generated legal assertion should be treated as safe unless tied to a real, inspectable source.

Security and confidentiality

For lawyer workflows, security is a product feature. User documents should not be used for model training by default. Reports should identify retained audit data, corpus versions, parser versions, and hashes so professional diligence can be reconstructed later.