Provisioned capacity sizing

Telemetry

For each provisioned deployment, UnitVane AI collects one day of per-minute utilization at a time and stores it rolled up to percentiles. Sizing uses a 30-day window by default; a deployment needs at least 7 days of data to be sized and 14 days to count toward reservation coverage.

Right-sizing

• The recommended size covers your sizing percentile of per-minute demand, converted with the model's tokens-per-minute per PTU.
• Deployments with spillover configured can size closer to their baseline, because peaks overflow to pay-as-you-go.
• Deployments marked latency-sensitive stay on PTU and are never recommended for pay-as-you-go.
• Deployments that would cost less on pay-as-you-go are shown with the break-even volume and how long a move stays cheaper.
• Recommendations only ever reduce cost. Increases are sent to the owning team for review.

Reservation coverage

Global, Data Zone and Regional capacity are separate pools with separate reservation products. In each pool, UnitVane AI compares reserved PTUs with PTUs deployed after right-sizing, and reports capacity committed beyond peak deployment and capacity deployed steadily but not reserved.

Automation

# Reservation automation, per pool. Purchases always wait for approval.apiVersion: ai.unitvane.com/v1kind: ReservationPolicymetadata:  name: data-zone-usspec:  pool: datazone/us  mode: renew-on-expiration     # off | purchase-monthly | renew-on-expiration  term: P1Y                     # P1M | P1Y | P3Y  scope: shared                 # shared | subscription | management-group  coverTo: deployed-after-rightsizing  approval:    required: true    approvers: [finops-leads]