Test Cloud admin guide

Tenant consumption enforcement is a per-service, per-tenant configuration that controls whether a service stops consuming units when it reaches the units allocated to that tenant.

In Test Cloud, licenses allocate a pool of units — such as Platform Units — to each tenant. Historically, services have continued to consume units past the tenant's allocated amount, drawing from the organization-level pool. This behavior meant that tenant allocations functioned as guidelines rather than hard limits, making cost control and budget governance difficult for organizations with multiple tenants.

Tenant consumption enforcement addresses this by letting Org Admins and Tenant Administrators configure, per service and per tenant, whether consumption stops at 100% of the allocated units or continues past the tenant limit into the organization pool.

Enforcement operates independently for each service on each tenant. When enforcement is enabled for a service, that service stops processing new units at 100% of the tenant's allocation. When enforcement is disabled, the service continues drawing from the organization-level pool after the tenant allocation is exhausted.

Each service is configured individually. There are no service groups — enforcement settings for one service do not affect other services on the same tenant.

When consumption continues past the tenant limit, the additional units are drawn from the organization-level pool. This provides a safety net for business-critical processes while still recording the consumption at the org level for reporting purposes.

Different services handle the enforcement boundary in different ways based on how they report consumption events to the licensing layer.

The licensing system receives consumption events when a process or activity completes, not mid-execution. This means that if a process completes and the consumption it reports pushes the tenant over 100%, that last event is accepted before enforcement halts subsequent consumption. Minor overconsumption at process boundaries is therefore possible even with enforcement enabled.

Enforcement activates at the next consumption event after the limit is hit, not mid-execution. If multiple processes are running when the limit is reached, all of them continue to completion and report their consumption — only new processes starting after that point are blocked. For example, if five processes are still running when the tenant limit is hit, all five continue to completion, because enforcement only activates after each process reports its consumption event at the end of its run. For services with larger individual consumption events, the overconsumption at the boundary may be proportionally larger.

Consumption events are reported at the end of a complete document processing job. When the tenant limit is reached, Document Understanding allows the current job to finish before stopping new jobs. As a result, a small amount of overconsumption may occur at job boundaries. The next job is blocked if enforcement is active and the allocation is at 100%. In cross-tenant scenarios where one tenant uses a Document Understanding model hosted on another tenant, consumption is always charged to the tenant hosting the model. To control this consumption through enforcement, enable enforcement on the model-hosting tenant.

The licensing system checks consumption at each individual LLM call within an agent run. When the tenant limit is reached with enforcement enabled, the agent may stop mid-run at the next LLM call boundary. The run does not complete.

Both services have a separate enforcement setting for each unit type, and the two settings interact. ScreenPlay runs consume ScreenPlay Runs first; when overconsumption is allowed and ScreenPlay Runs are exhausted, ScreenPlay draws from the available Platform Units (Unified Pricing model). Healing Agent runs consume Heals first; when overconsumption is allowed and Heals are exhausted, Healing Agent draws from the available Agent Units (Flex pricing model).

When only the overflow unit is allocated to the tenant (Platform Units for ScreenPlay; Agent Units for Healing Agent):

When both unit types are allocated to the tenant:

For ScreenPlay: primary unit = ScreenPlay Runs, overflow unit = Platform Units. For Healing Agent: primary unit = Heals, overflow unit = Agent Units.

Tenant-level enforcement is not yet implemented for these services. They appear unavailable in the configuration interface and always continue consuming from the organization pool, regardless of the enforcement setting on the tenant.

Consumption notifications are sent as services approach and reach the tenant allocation, regardless of whether enforcement is enabled or disabled. Notifications are sent at the following thresholds for tenant-level consumption:

All Organization Administrators and Tenant Administrators receive these notifications by default. For information on all licensing notification events, see Licensing notifications.

When tenant consumption enforcement becomes available in your organization, the default setting depends on your organization's current consumption state.

Organizations in the second category should audit their tenant allocations before enabling enforcement, to avoid blocking production processes that currently rely on the organization pool.