Beyond search, the platform reads the collection the way a first-pass reviewer would - and records its findings as structured, filterable data on every document. The result is a coded evidence log for the whole matter, available from day one instead of week six.
What Gets Coded
- A concise summary and issue tags drawn from the matter's own issue taxonomy, for both claims and counterclaims.
- Bearing - does this document help or hurt your position - with a stated confidence level.
- Key quotes, verbatim and attributed, translated when the source is in another language.
- People and companies mentioned, with their roles, and every dollar amount.
- Dated facts that feed the case chronology.
- Privilege flags with a stated basis - attorney-client or work product - and the attorneys involved.
- Party admissions and hot document flags for the material that decides cases.
- A mapping to the specific paragraphs of the complaint each document speaks to.
Discipline That Keeps It Honest
Any claim that a document helps, hurts, or is hot must be justified by a verbatim quote from the document itself - no quote, no call. And all AI output is flagged as derived attorney work product: it is searchable alongside the evidence, but it can never be produced as if it were native ESI. The record stays the record.
Cost Under Control
A flat, single-model pass over a large collection wastes most of its budget coding newsletters. The platform runs a tiered funnel instead: a free bulk screen removes obvious noise, a fast model triages what remains, and a deep model codes only what matters. Budget guards require explicit confirmation before any run that exceeds a configured limit, live cost estimates are shown before you commit, and completed work is never re-billed - every batch is journaled so results can be recovered and re-applied without paying twice.
Works Together With
- Hybrid SearchSemantic and keyword search fused and reranked, with the custodian, date, and issue filters that discovery questions actually take.
- Case IntelligenceA cited case chat, chronology, hot and adverse document queues, truth table, entity graph, and per-claim strategy pages.
- Bates ProductionReview decisions to a served, Bates-numbered package with a fail-closed privilege gate and a verifiable manifest.
Frequently Asked Questions
Does AI coding replace attorney review?
No - it gives attorneys a structured first pass over the entire collection, so human review time goes where it matters. Every AI call is backed by a verbatim quote and a stated confidence level, and attorneys make every decision that counts, from review calls to production.
How do you keep the AI honest?
Two disciplines. First, any claim that a document helps, hurts, or is hot must be justified by a verbatim quote from the document itself. Second, all AI output is flagged as derived attorney work product: searchable alongside the evidence, but never producible as if it were native ESI.
What does coding a large collection cost?
Coding runs through a tiered funnel - a free bulk screen removes obvious noise, a fast model triages what remains, and a deep model codes only what matters - which cuts cost to a fraction of a flat single-pass approach. Live estimates are shown before you commit, budget guards require explicit confirmation above a configured limit, and completed work is never re-billed.
How does coding handle privilege?
Privilege is coded as a first-class field with a stated basis - attorney-client or work product - and the attorneys involved. Those flags feed the draft privilege log and the fail-closed privilege gate at production time, so a flagged document cannot slip into a production by accident.