Litigation Document Review
The review room, industrialized — responsiveness, privilege, and the hot documents, at corpus scale.
Litigation document review is the assessment of discovery corpora against the case's legal questions: which documents are responsive to the requests, which are privileged, which bear on which issues, and which are the hot documents — the emails and memos that will anchor depositions, motions, and settlement calculus. It is historically litigation's largest cost center: rooms (then screens) of contract attorneys reading at hundreds of documents per day each, for matters whose corpora run to millions — the economics that made review the first legal task to industrialize and the first to accept machine assistance as defensible.
The technology arc runs from search terms through TAR to language models. Keyword culling narrowed corpora crudely (and litigably — term negotiation is its own motion practice). Technology-assisted review brought statistical classifiers trained on attorney coding, with validation protocols (recall estimation on samples) that courts accepted and, in some quarters, came to prefer over exhausted human linearity. The LLM generation reads rather than correlates: review instructions applied as instructions, responsiveness assessed against the actual request language, privilege recognized in meaning (advice sought, counsel consulted) rather than lawyer-name proximity, and every call explained — the explanation accelerating the human QC layer that defensibility still demands. Issue coding, chronology building, and hot-document surfacing ride the same capability: the corpus not just sorted but read, with the narrative threads assembled.
The operating disciplines are litigation-specific: privilege calls human-verified before production (the asymmetry of a privileged document produced justifies the cost), validation statistics disclosed and negotiated, protocols and model versions documented against challenge, and the review platform's audit trail treated as work product that opposing counsel may probe. What changes with each technology generation is the ratio — attorney hours per thousand documents falling — while what persists is the structure: machines propose, sampling verifies, and the certification a lawyer signs still means a lawyer stands behind the production.
A review team that reads a million documents without fatigue — finding the responsive, flagging the privileged.
Before review comes the grind: collecting, extracting, deduplicating, and staging a million files for the reviewers.
The argument, its authorities, and its weak joints — briefs read structurally, citations checked.
Proof Perimeter runs document AI inside your own perimeter — with a provenance record on every field.
Book a demo