Five decisions. One of them sent.
The message that didn't send is the product. These are the five canonical outcomes, each with the full receipt behind it. Open any one — it is designed to survive being screenshotted into a compliance review.
Lab result follow-up
Lab results were ready and a follow-up was due. Every hard stop cleared, the judgment engine proposed a send, all three governance gates passed, and Meta accepted the dispatch.
- AI judgment
- invoked
- Dispatch
- Accepted by provider
- Policy
- uhc_labs_followup.v2
Re-engagement check-in
The judgment engine proposed a check-in after six quiet weeks. Regent held it: the care team had spoken with the subject the previous day. The hold is terminal — nothing is queued and nothing will fire later on its own.
- AI judgment
- invoked
- Dispatch
- No dispatch attempted
- Policy
- uhc_reengagement.v3
Post-discharge engagement
A routine post-discharge engagement was requested. A crisis flag was already present in the accepted snapshot, so the pre-agent hard stop resolved the operation before the judgment engine ran. No outreach was proposed and none was sent.
- AI judgment
- NOT RUN — pre-agent hard stop
- Dispatch
- No dispatch attempted
- Policy
- uhc_reengagement.v3
Annual check-in
An annual check-in was pushed at 22:14 local time, outside the configured send window. The operation ended as DEFER_REPUSH_REQUIRED. Nothing is queued and nothing fires when the window opens — the customer re-pushes.
- AI judgment
- invoked
- Dispatch
- No dispatch attempted
- Policy
- uhc_reengagement.v3
Appointment reminder
Every gate passed and dispatch was authorized, but the provider send response was lost in transit. No provider message id came back, so acceptance is unresolved. Regent does not retry blind — the operation waits for a provider-backed identifier or an operator.
- AI judgment
- invoked
- Dispatch
- Acceptance unresolved
- Policy
- uhc_labs_followup.v2
A governed operating day is mostly not-sending.
The console holds the whole synthetic dataset — suppressions, provider rejections, a controlled error, a processing-promise defer. Each one is an ordinary outcome with an ordinary receipt.
- SUPPRESSEDOutcome: suppressed, not sentOverdue lab reminder
- PROVIDER REJECTEDOutcome: rejected by the provider, not deliveredLab result follow-up
- SENTOutcome: sentCare-plan summary
- HOLD · TerminalOutcome: terminal hold, not sentRe-engagement check-in
- ESCALATEDOutcome: escalated to a named human, not sentMedication adherence nudge
- DEFERREDOutcome: deferred, customer re-push required, not queuedLab result follow-up
- ESCALATEDOutcome: escalated to a named human, not sentOvernight welfare check
- ESCALATEDOutcome: escalated to a named human, not sentOvernight welfare check
- ERROROutcome: error, controlled failureRe-engagement check-in
Synthetic demonstration dataset. Not production telemetry.
Change the context and watch the outcome change.
The Governance Playground runs the same gates over the same fixtures. Set a crisis flag and the judgment engine stops being invoked. Move the clock outside the send window and the operation defers.