Commissioning preview

Production readiness

Do not confuse a working application with a commissioned clinical tool.

The code can be deployed for internal evaluation. Clinical use requires the following evidence, governance, privacy, security, and operations gates to be completed and documented.

Open

Evidence commissioning

  • Complete licensed database searches
  • Document dual screening and extraction
  • Finish formal RoB, GRADE, and AGREE II
  • Reconcile exact current labels and access dates
  • Obtain five named external specialist sign-offs
  • Freeze a commissioned release and change log
Open

Clinical validation

  • Write representative gold-standard cases
  • Blind-review retrieval relevance and output fidelity
  • Test rare, conflicting, and insufficient-evidence cases
  • Validate red-flag bypass and failure states
  • Define acceptance thresholds and rollback rules
  • Approve clinician onboarding and competency checks
Open

Privacy & security

  • Validate that raw quick-case text never leaves browser memory
  • Red-team quick-case parsing, ambiguity, and identifier rejection
  • Verify OpenAI Zero Data Retention project controls
  • Complete vendor, BAA/DPA, legal, and privacy review
  • Place the site behind Cloudflare Access
  • Verify no body logging across every layer
  • Run threat model, penetration test, and secret review
  • Approve incident-response and breach procedures
Open

Regulatory & quality

  • Complete FDA CDS/device classification assessment
  • Complete state professional-practice review
  • Establish a quality management and change-control owner
  • Define complaint, correction, and recall workflows
  • Approve labeling, intended use, and limitations
  • Validate accessibility against WCAG 2.2 AA
Open

Living evidence

  • Assign surveillance owners and cadence
  • Automate source discovery without auto-publishing
  • Require human evidence appraisal and approval
  • Publish cutoff, version, amendments, and corrections
  • Trigger urgent review for regulator safety actions
  • Test rollback to the previous evidence release
Open

Operations

  • Set rate, spend, timeout, and concurrency limits
  • Define downtime and API refusal behavior
  • Monitor service health without collecting case bodies
  • Provide a staffed clinician feedback channel
  • Audit access and offboarding
  • Schedule prompt, model, and evidence regression tests
Launch rule

If a required control cannot be verified, fail closed or keep the affected function in internal demonstration mode. No disclaimer substitutes for a missing safety control.