Northwind Health Vendor DPA
Four expected deviations, including a 120-hour incident-notice term.
Playbook-based contract review
Six firm rules tested against the vendor paper. Every released finding must have source evidence.
Independent review protocol
These roles have different objectives and permissions—not theatrical personalities. The Evidence Auditor is required to oppose unsupported or overstated conclusions before anything reaches an attorney.
Recorded disagreement
Configuration
Firm standards convert legal judgment into repeatable review instructions. A practice leader approves every change before agents use it.
Decision record
Golden matter
Clause extraction, playbook comparison, source verification, and draft language.
Risk acceptance, final redlines, client advice, and external communication.
Synthetic demonstration data. Matter-level access and retention controls required for a client pilot.
Synthetic benchmark results from an independent prototype. A real pilot requires client-attorney-approved ground truth.
Portfolio intelligence
Three vendor DPAs tested against the same approved playbook. Open any exception to verify its source.
Playbook comparison
Post-signature operations
Turn approved contract terms into owned actions, reminders, and escalation decisions.
Northwind Health Vendor DPA
Administrator workspace
Select approved sources, manage versions, and keep the evaluation answer key separate from the user workflow.
Approved sources
Four expected deviations, including a 120-hour incident-notice term.
Mostly compliant; the 120-day backup period tests calibrated severity.
No fixed incident deadline tests uncertainty and safe escalation.
Six attorney-approved rules with severity and escalation guidance.
Permitted use, prohibited autonomy, data handling, logs, and change control.
Expected findings, pages, severity, challenge behavior, and release thresholds.
System scope, NIST AI RMF crosswalk, Illinois applicability, controls, evidence, and release criteria.
Release controls
Administrator assurance workspace
Translate responsible AI commitments into accountable owners, operating controls, measurable evidence, and attorney-approved release decisions.
Operating model
Purpose, users, intended benefit, risk appetite, prohibited uses, and accountable executive.
Evidence: AI System Card · Use-case approvalNamed operator and supervisor, approved sources, training, access, workflow, and change control.
Evidence: Role register · Playbook v3.2Confidentiality, hallucination, bias, privilege, vendor, security, escalation, and stop controls.
Evidence: Risk register · Challenge ledgerCritical recall, citation accuracy, attorney acceptance, cycle time, overrides, and adoption outcomes.
Evidence: Golden matter scorecardHuman accountability, protected client interests, transparent use, and uses the firm refuses to automate.
Evidence: Responsible AI policyGovernance record
Control crosswalk
Illinois applicability
Competence, confidentiality, supervision, candor, sensitive-information protection, and attorney responsibility for final work.
Assess when production documents contain covered personal information; apply security, retention, and incident procedures.
No biometric processing or employment decision-making is permitted in this contract-review pilot. Reassess before expanding the use case.
Runtime assurance exercise
Evidence Auditor finds no approved authority supporting the claim.
Finding is withheld and cannot reach the Drafting Agent.
Matter attorney receives the issue and missing-evidence reason.
Claim, challenge, disposition, configuration, and owner enter the audit trail.
Only an attorney-approved authority or corrected playbook can release a replacement finding.
Self guided lawyer pilot
Review one fictional agreement, verify the evidence, inspect agent disagreements, make attorney decisions, and tell us what would or would not work in practice.
First session
Return to Contract review and choose Northwind Health. It contains Critical, High, and Medium deviations.
Open the agreement and approved playbook. Keep them available so you can verify quotations and rules.
Observe the separate Intake, Clause, Risk, Evidence Audit, and Drafting outcomes.
Check the contract quotation, PDF page, playbook rule, severity, rationale, and proposed revision.
Open Agent challenges. Confirm unsupported or overstated conclusions are opposed and withheld from drafting.
Accept, reject, or escalate each proposal. Check that your actions appear in the audit trail.
Open Evaluation only after your review. Record misses, false positives, wrong pages, severity errors, or unsafe claims.
Open Governance & assurance to review the System Card, control crosswalk, Illinois applicability, release criteria, and runtime assurance exercise.
Scenario design
Tests recall, exact evidence, severity calibration, and useful redlining across several issues.
Tests precision and restraint. A trustworthy system should not manufacture findings to fill a queue.
Tests uncertainty handling. The agents should identify the gap without inventing a deadline or legal rule.
Small and medium law firms
What good looks like
Did the workflow surface every material issue you expected?
Did it avoid turning compliant language into a finding?
Did every quotation and page citation match the source?
Was every conclusion tied to the correct approved rule?
Were severity, escalation, and fallback language proportionate?
Were unsupported claims challenged and attorney authority preserved?
Could a lawyer understand the next action without additional explanation?
What integration, customization, or control would your firm require?