eDiscovery

Discovery volume invites automation, and automation invites a privilege mistake at scale. One clumsy prompt can move a sealed document into a request, and one confident answer can read like a legal conclusion. This starter narrows both paths and expects counsel to keep the judgment.

Discovery teams reach for this starter when review sets and privilege logs already touch a model. Case and docket numbers in a prompt are masked, so a request does not carry a live matter reference upstream. The block list refuses a prompt marked privileged and confidential, a prompt naming a sealed filing, or one stating a settlement amount.

History capture is set to metadata only, and it includes calls that policy blocked. Capture does not fail open, so a capture failure does not quietly let traffic pass. That gives your team a call level trace, and it is not an evidence archive.

Output that directs someone to file this lawsuit, or to sign this settlement, is refused on the way back. The privilege stage has an empty phrase list, and it cannot decide what is privileged in law. Citation checking marks an unverified reference and returns the text anyway, which is the default behavior. An answer must score 0.90 overall, 0.78 on relevance and 0.80 on accuracy, or it is blocked.

Decisions about collection, preservation and production stay with counsel and sit outside this starter. The oversight entry is empty, so a flagged finding does not reach a reviewer until you configure that step. Audit marking allows and notes, and it creates no storage of its own. Among the legal starters here, this one is aimed at review volume rather than at drafting.