Evidence Intelligence

Evidence Intelligence ingests every document in a matter — pleadings, discovery productions, transcripts, medical records, exhibits, notes — and produces a structured index of the factual assertions those documents contain. Each extracted fact is anchored to the specific passage it came from so an attorney can verify it in one click. It is the foundation on which every downstream engine (timeline, witness, constitutional, motion, report) is built.

1. 1
   Step 1

   Documents are extracted and normalized. Scanned PDFs and images pass through OCR so that photographed reports and handwritten notes become searchable text.
2. 2
   Step 2

   The corpus is chunked and indexed. A retrieval layer maintains embeddings for every passage so the model can look up support before writing anything.
3. 3
   Step 3

   Extraction engines identify entities, events, statements, and factual claims, tagging each with the document, page, and paragraph it appeared in.
4. 4
   Step 4

   An evidence gate suppresses any claim the retrieval layer cannot ground back to a source passage. Suppressions are logged in the pipeline ledger for review.
5. 5
   Step 5

   Findings are written to your workspace with citation metadata attached so downstream engines can reuse them without regenerating.

1. 1
   Create a case

   Give the matter a name and case type.
2. 2
   Upload the corpus

   Drag in the full production. OCR runs automatically on scans.
3. 3
   Run analysis

   Evidence Intelligence extracts and indexes the record.
4. 4
   Review findings

   Open the Evidence panel and verify each cited claim.

• Upload the entire production, not curated excerpts — downstream contradiction detection depends on it.
• Label documents clearly (bates ranges, produced-by) so citations resolve unambiguously in a filing.
• Rerun extraction after supplemental productions so the evidence index stays current.
• Review the evidence panel before moving to motion drafting — the gate is stricter earlier than later.

• Verify each cited claim against the source before relying on it.
• Redact privileged material before uploading to a shared workspace.
• Confirm that OCR-derived text matches the original where the extraction is close-call.
• Retain final judgment on what constitutes admissible or material evidence.

• OCR accuracy on very low-quality scans or dense handwriting may require attorney correction.
• Extraction quality depends on document structure — heavily tabular exhibits can require secondary review.
• Attribution of statements to a specific speaker is inferred from context and is not infallible.