← Agent Registry

Clinical Data Quality Agent

Identifies data query backlogs, deviations, missing data, and data integrity signals.

Planned
Status
Planned
Owner
Data Management
Version

Approved Data Connections

Data access is permission-scoped and governed by Agentic OS policies.

  • Patient-level PHI
  • Unapproved clinical datasets

Agent Permissions

Can analyze · Can prepare · Requires approval · Cannot execute autonomously.

Allowed
  • Retrieve approved data
  • Analyze operational status
  • Calculate risk
  • Generate explanations
  • Recommend actions
  • Prepare actions
Human approval required
  • External communications
  • Operational escalations
  • Workflow changes
  • Creating external tasks
Not allowed
  • Modify clinical source data
  • Access unauthorized datasets
  • Access patient-level PHI
  • Execute high-impact actions autonomously

Autonomy Level

Level 1 — Retrieve
Level 2 — Recommend
Level 3 — Prepare
Level 4 — Human-Approved Action
Current level
Level 5 — Autonomous Action
Not enabled

The agent may retrieve approved data, analyze operational risks, recommend interventions, and prepare actions. Consequential actions require explicit human approval.

AI Configuration

Model
Enterprise-Approved LLM
Reasoning mode
Hybrid
Deterministic logic
  • Permissions
  • Data validation
  • Risk thresholds
  • Workflow state
  • Fixed business rules
LLM reasoning
  • Intent interpretation
  • Cross-source synthesis
  • Contextual explanation
  • Recommendation generation

Principle: use deterministic automation where possible and AI reasoning where it adds contextual value.

Agent Evaluation

Synthetic Prototype Metrics — not production outcomes.

Evaluation suite Site Readiness Eval v1.0
Last evaluation Aug 2026
Passed
Risk identification accuracy
92%
False-positive rate
8%
Recommendation approval rate
67%
Human override rate
33%
Successful issue resolution rate
1/2
Average time to resolution
4.2 days

Outcomes feed the shared Agent Performance experience.

Audit & Traceability

  • User request
  • Agent identity and version
  • Data sources accessed
  • Risk assessment
  • Recommendation
  • Proposed action
  • Human approval / rejection
  • Executed action
  • Outcome
  • Evaluation feedback

All agent activity is traceable through the shared Agentic OS audit framework — visible in the Audit Trail. Approvals appear in Approvals.