AI Transparency

What our AI does, what it does not do, and how you stay in control.

Extraction vs. interpretation

Extraction pulls facts out of what you upload: text, names, dates, times, locations, senders, recipients, attachments, visible metadata, and reply relationships.

Interpretation suggests possible groupings, recurring documented themes, documentation gaps, and clarification questions. Interpretation is always shown as a suggestion, never as a fact.

You decide

Every AI suggestion supports Confirm, Edit, Reject, Unsure, and Review later. Rejected suggestions never appear in an export.

Explanations, not hidden reasoning

Every AI output shows what PatternProof did, why, which sources it used, what remains uncertain, and what you can do next. We do not expose private model reasoning or chain-of-thought.

Provenance we store with each AI output

Source record IDs the suggestion was drawn from
Kind of AI activity (extraction or interpretation)
Model identifier and version, and instruction version
Structured output, timestamp, and your response
Revision and export history

What AI does not do

PatternProof does not diagnose an alleged abuser, predict future violence, calculate the probability that abuse occurred, or replace professional judgment. It helps you preserve, organize, and describe what you already have.

This page describes controls that are implemented in PatternProof today. If a description does not match your experience, please tell us. Related: Privacy · Safety · Evidence integrity · AI transparency · Professional access