> ## Documentation Index
> Fetch the complete documentation index at: https://headgate.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Fleet policies

> Rate limits, fairness, concurrency, queue controls, priority, and quarantine.

Policies read by the admission gate are stored fleet-wide and writable through the control
API. Worker-local configuration cannot silently change shared behavior.

<CardGroup cols={2}>
  <Card title="Tenant fairness" icon="scale">
    Work-conserving deficit service across `partition_key` values, including quiet groups under a flood.
  </Card>

  <Card title="Rate classes" icon="gauge">
    Shared token limit, window, burst, job weight, and an operator pause switch.
  </Card>

  <Card title="Concurrency ceilings" icon="layers-3">
    Global in-flight limits enforced during the same admission operation.
  </Card>

  <Card title="Quarantine" icon="shield-alert">
    Repeated crash fingerprints are parked visibly until an operator releases them.
  </Card>
</CardGroup>

## Ordering rules

Queue weight selects which queue receives service. Job priority orders work only within a
partition. Neither field overrides the other.

## Explainability

`GET /jobs/{id}/admission` reports whether a job is admissible, which policy blocks it,
and whether the condition is expected to clear. This is a first-class operational API,
not a debug-only implementation detail.

`blocked_by: null` does not mean an unknown policy. It means no admission policy is
blocking the job. The job may already be running, may be pending explicit promotion, or
may have reached a terminal state where admission no longer applies. Likewise,
`estimated_admission_ms: null` means no time estimate is available; the blocking-policy
name determines whether capacity can clear naturally or an operator must act.