# Capacity model

## Workload assumptions

- Arrival rate, burst shape, concurrency, and seasonality:
- Input/output tokens, context reuse, tool latency, and agent steps:
- Model/hardware/deployment version:

## Service demand

- Prefill and decode demand:
- KV/cache/storage demand:
- Tool/database/network demand:
- Human-review arrival and service time:

## Queues and limits

- Admission, rate, concurrency, and budget limits:
- Expected utilization and waiting time:
- Overload and backpressure behavior:

## Sensitivity and decision

- Base, peak, and failure scenarios:
- Dominant assumptions and break-even points:
- Capacity owner and next measurement: