WORKED EXAMPLE · FICTIONAL

The AI said cancelled. The appointment stayed scheduled.

Inspect six linked fictional artifacts, a bounded conclusion, and the owner-specific follow-ups needed to resolve the discrepancy.

Access Red Team · 15 September 2026 · Free to use, no account required

FICTIONAL WORKED EXAMPLE · No real patient, vendor deployment or executed system test

Six artifacts. One synthetic cancellation request.

Packet manifest: SYN-CANCEL-041 · 15 September 2026 · observation cutoff 10:05 UTC. The caller requested cancellation of APPT-SYN-041. The six excerpts below are authored teaching artifacts, explicitly correlated by this manifest to that same synthetic request and appointment. Their identifiers are fictional labels, not vendor-native formats.

For this exercise, A5 is the supplied scheduling-state record. In a real review, establish its authority, freshness and correlation with the scheduling owner. No later recovery record or causal trace is supplied.

Download the complete fictional packet (TXT) ↓

A1 · 10:03 UTC

Call summary

CALL-SYN-041 · SYN-CANCEL-041

Call connected and ended normally.

A2 · 10:02 UTC

Conversation

CONV-SYN-041 · SYN-CANCEL-041

AI told the caller: “Your appointment has been cancelled.”

A3 · 10:02 UTC

Integration summary

MSG-SYN-041 · SYN-CANCEL-041

A callback request was accepted for processing.

A4 · 10:03 UTC

CRM record

CASE-SYN-041 · SYN-CANCEL-041

Case closed because it was sent to another queue.

A5 · 10:05 UTC

Scheduling record

APPT-SYN-041 · SYN-CANCEL-041

Appointment remained scheduled at the observation time.

A6 · 10:05 UTC

Receiving-process evidence

RECEIVE-SYN-041 · SYN-CANCEL-041

Acceptance by a receiving team was not established in the supplied packet.

Write only the conclusion the packet supports.

The supplied evidence does not support the cancellation confirmation at the observed time. It shows a callback submission and administrative case closure while the appointment remains scheduled. Receiving-team acceptance remains unconfirmed. The packet does not establish which component caused the discrepancy or whether later recovery occurred.

Why this wording matters

  1. “The supplied evidence” limits the conclusion to this packet. It does not claim every system or record was examined.
  2. “At the observed time” ties the finding to the 10:05 UTC snapshot. It leaves later recovery open.
  3. “Callback submission and administrative case closure” reconciles A3 and A4. Neither says an appointment cancellation completed.
  4. “Receiving-team acceptance remains unconfirmed” reflects A6. A missing acceptance record does not establish that nobody acted.
  5. No cause is assigned. The packet does not distinguish configuration, application behavior, integration behavior, timing or operational causes. Investigate before presenting any as established.

The caller's cancellation confirmation is unsupported by the supplied outcome evidence. Whether this service should cancel directly or request staff follow-up is a separate scope question that the operational owner must answer.

Make each unknown assignable.

Follow-up questions for the same request
Responsible roleNext confirmation
AI application ownerWhich confirmed milestone supported the statement to the caller?
Integration ownerWhat did the acknowledgment establish?
CRM ownerWhat does this closure reason mean?
Scheduling ownerWhich record establishes the appointment’s state?
Receiving-team ownerWho owns the request until someone accepts it?
Operational ownerIs the service supposed to cancel appointments directly or request staff follow-up?

Ask contact-center administration to confirm the call event's local meaning, and access-policy ownership to identify the permitted cancellation and delegation rules. The packet does not establish whether the caller had authority to cancel.

Record the answers and their evidence references in the completed milestone worksheet. Unresolved rows are deliberate: they identify the next work rather than inventing an answer.

Start with the accountable workflow owner.

Agree the requested outcome, observation window and decision first. Involve the contributors needed for that scope; one person may hold several roles. Record names in your own approved workspace.

  • Contact-center administration: call events, queue definitions and caller communication.
  • AI application ownership: the milestone used to generate a completion statement.
  • CRM ownership: case states, closure reasons and transfer behavior.
  • Integration ownership: message correlation and acknowledgment semantics.
  • EHR/scheduling ownership: the authoritative appointment record and its observation time.
  • Receiving operations: acceptance, due windows, recovery and closure outcomes.
  • Access-policy ownership: who may request and perform the action, and under which conditions.

The accountable workflow owner coordinates unresolved questions across these roles. A named team is not evidence that someone accepted a particular request.

INDEPENDENT WORKFLOW ASSESSMENT

One patient request. Several systems. Who can prove the outcome?

Patient-access automation can cross telephony, AI, CRM, EHR, and the teams responsible for follow-up. Understanding the result requires evidence from the complete workflow.

Access Red Team provides external, independent red teaming of your configured workflow. We execute agreed tests, validate findings against recorded outcomes, and retest agreed corrections.

Your team receives evidence showing what was tested, what happened, where responsibility sits, and what remains unresolved.

Executed testing. Validated findings. Retested corrections. Scope and limitations stated.

Validation applies to findings within the agreed scope. Unconfirmed observations remain labeled as such, and retesting covers the agreed corrections and cases.

Scope an independent workflow assessment

NIST’s AI RMF 1.0 describes independent review as a way to strengthen testing and reduce internal bias or potential conflicts of interest. This supports independent review; external status alone does not guarantee better work. NIST AI RMF, Measure.