CONTEXT(CLONE)
Sign in

First-pass relevance and privilege review for document disclosure

For: Litigation associate or disclosure manager facing a court or regulator deadline

Pattern: Map, verify, reduceNeeds scaleDesigned for 50 to 2,000 agents
Get early accessSign in to use

The pain today

A very large set of emails and attachments must be tagged for relevance to the pleaded issues and screened for privilege. Reviewers tire and tag inconsistently, and a privileged document produced by mistake is hard to claw back.

The ask

“I attached the list of issues from the pleadings, the names of our lawyers and advisers, and the collected emails and attachments. For each document, tell me which issues it is relevant to and whether it may be privileged, with the passage and reason, and list the ones you are unsure about.”

Plain words, as you would say it to a colleague. Edit it to fit your case before you send it.

What you attach or connect

  • Collected emails and attachments
  • List of issues from the pleadings
  • Names and roles of lawyers and advisers
  • Review protocol with tagging definitions

The unit of work

One worker task per one document family (email with attachments).

Why a swarm fits

Each document family is tagged on its own against a short issue list and protocol. The volume is the problem, and an independent second look at every privilege call is worth far more than its cost.

Not for

A small disclosure a paralegal can read in days, or a matter where the court has already approved your review platform's trained classifier.

The decision tree

6 typed decisions, each with an action for every answer

At fixed moments in a run, the engine puts one narrow question to a decision model. The decision model never writes text: it answers yes or no with a probability, picks from listed options, or gives a score, about a small slice of the material. The engine then does exactly what this tree says, which is what makes the run auditable. The thresholds are the template's design values, not measured results.

  1. Planner, while planning

    Scope checkYes or no, with a probability

    Before work starts on a unit

    Does this document family fall inside the date range and custodians set by the review protocol?

    Sees only: The family's metadata and the protocol's scope paragraph

    Why: Out-of-protocol documents are excluded before any reading cost.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.30 up to 0.60thenAccept
    • No: below 0.30thenSkip this unit

  2. Before workers, before a task runs

    Small worker or strong modelYes or no, with a probability

    Before a task runs

    Is this a short plain email with no attachment, no lawyer among the participants and no forwarded chain?

    Sees only: The family's participants, length and attachment list

    Why: Long chains and lawyer involvement get a stronger first reader.

    • Yes: 0.60 or higherthenAccept
    • Unsure: 0.40 up to 0.60thenEscalate to a strong model
    • No: below 0.40thenEscalate to a strong model

  3. After workers, the judge checks

    Evidence checkYes or no, with a probability

    After a worker answers

    Does the quoted passage discuss the facts of the pleaded issue it was tagged to, rather than merely share a keyword with it?

    Sees only: The quoted passage and the one pleaded issue

    Why: Keyword-level tagging floods the lawyers' queue.

    • Yes: 0.85 or higherthenAccept
    • Unsure: 0.50 up to 0.85thenEscalate to a strong model
    • No: below 0.50thenReject and retry

  4. Evidence checkA choice among options

    After a worker answers

    From the participants and the quoted passage, what is the privilege position of this document?

    Sees only: The participants with roles and the quoted passage

    Why: A missed privilege call is the one error that cannot be taken back.

    • Lawyer involved, advice sought or giventhenAsk a person
    • No lawyer and no advice discussedthenAccept
    • Lawyer copied, purpose unclearthenAsk a person

  5. Reconciler, while merging

    Conflict checkYes or no, with a probability

    While reconciling

    Do two documents of the same thread, or two near-duplicates, carry different relevance or privilege tags?

    Sees only: The two documents' tags, passages and similarity

    Why: Inconsistent tags across a thread are what opposing counsel notices.

    • Yes: 0.60 or higherthenMark unresolved
    • Unsure: 0.30 up to 0.60thenMark unresolved
    • No: below 0.30thenAccept

  6. Accountable person, before anything is settled

    Person decidesYes or no, with a probability

    Before anything is reported as settled

    Does this family involve a listed lawyer or adviser, or carry a privilege flag from either the worker or the judge?

    Sees only: The family's participants and both sets of tags

    Why: Every privilege call is made by a qualified lawyer.

    Accountable: A qualified lawyer makes every privilege call and signs the disclosure statement; the swarm only proposes tags with reasons.

    • Yes: 0.20 or higherthenAsk a person
    • Unsure: 0.05 up to 0.20thenAsk a person
    • No: below 0.05thenAccept

The fleet: who does what

Model tiers by role, not brands: you choose the models. Strong reasoning models plan and reconcile, small fast models do the wide work, and the judge is a decision model from a different family, so it does not share the workers' blind spots.

  1. Planner

    A strong reasoning model turns the pleadings and protocol into tagging questions with examples of each tag.

    Decisions here:1. Scope check

  2. Workers

    Small fast workers from an open-weight family tag one document family for issue relevance and possible privilege.

    Designed for 50 to 2,000 agents, one worker task per one document family (email with attachments). Each worker receives only its own unit.

    Decisions here:2. Small worker or strong model

  3. Judge, from a different model family

    A decision model from a different family re-decides every privilege flag and a sample of relevance tags against the quoted passage.

    Decisions here:3. Evidence check4. Evidence check

  4. Reconciler

    A strong reasoning model checks consistency across threads and duplicates and assembles the review queues for lawyers.

    Decisions here:5. Conflict check

  5. Accountable person

    A qualified lawyer makes every privilege call and signs the disclosure statement; the swarm only proposes tags with reasons.

    Decisions here:6. Person decides

Checked before anything is accepted

  • Every tag cites the passage and the issue or privilege ground it relies on
  • Near-duplicates and the same thread must carry consistent tags or be queued
  • Any document involving a listed lawyer or adviser goes to human review regardless of tag
  • Documents the workers and judge disagree on are never auto-tagged

What comes back

  • Proposed tags per document with passage and reason
  • Privilege review queue for lawyers
  • Disagreement and low-confidence queue
  • Consistency report across threads and duplicates

What to measure

  • Tags overturned in lawyer quality-control samples
  • Privileged documents caught by the judge after a worker missed them
  • Lawyer hours per reviewed document

Names of measures only. No result is claimed for this template.

Templates open in the workspace chat with the ask filled in. Nothing runs until you send it.

Get early accessSign in to use

More in Legal and contracts

Find where signed contracts deviate from the contract playbook

For: General counsel or legal operations lead who inherited a contract portfolio

Nobody knows which signed agreements carry uncapped liability, odd governing law or weak termination rights.

Pattern: Map, verify, reduceRuns today5 decisionsDesigned for 4 to 250 agents

Settle assignment and change-of-control readings before a transfer

For: In-house or deal counsel planning an intra-group reorganisation, carve-out or sale

Whether consent is needed turns on wording: assignment versus transfer by operation of law, direct versus indirect control, affiliate carve-outs.

Pattern: Cross-examinationNeeds live models5 decisionsDesigned for 6 to 200 agents

Draft and rank fallback wording for contested clauses

For: Commercial counsel refreshing a template or stuck on a contested set of clauses

Counsel drafts one or two alternatives under time pressure and sends the first that reads well.

Pattern: TournamentNeeds live models6 decisionsDesigned for 8 to 150 agents