How it works
Six steps, each of which leaves a checkable trace. Nothing in the report is asserted without a source reference.
The pipeline
1. Upload your documents
PDF, DOCX, TXT and images. Files are validated and stored in a private bucket, and each one is hashed on the server so its identity is fixed from the moment it arrives.
2. Text is extracted with provenance
PDFs are read page by page; scans and images go through OCR. Text is split deterministically into chunks that record page number, section, character range, extraction method and extraction confidence.
3. Staged analysis
Entities, events, key facts, contradictions and uncertainties are extracted in separate passes, then composed into a report structure. Evidence text is fenced so instructions hidden inside a document cannot influence the analysis.
4. Citations are validated, then certified
Every quoted excerpt is matched verbatim against the stored chunk it claims to come from; anything unverifiable is dropped. The evidence manifest, report document and PDF are hashed and certified as a Certified Execution Record.
5. You receive the report
A professional PDF with an executive summary, chronological timeline, findings, contradictions, uncertainties and a full source manifest — plus a verification page you can share.
6. Sources are deleted
Your source files and the extracted text are removed automatically, typically within 24 hours. Hashes and certification metadata remain, so the report stays verifiable after your documents are gone.
Why verification matters
Anyone can produce an AI summary of a document. The difficulty is showing a third party that the summary they are reading is the one that was actually produced from those exact documents, unchanged. That is what the Certified Execution Record does — and it is proof of process, not proof of truth. Read the full scope and limitations.