> ## Documentation Index > Fetch the complete documentation index at: https://docs.doblier.io/llms.txt > Use this file to discover all available pages before exploring further. # Hardware Sizing Guide > How ORCA deployments are sized — and how to get sizing for your estate. ORCA deploys in three reference tiers. Sizing is driven by four measurable inputs from your environment: raw events/day, peak events/sec, dedup-state size (the dominant memory driver), and the size of your dependency graph. | Tier | Profile | Typical estate | | ------------------ | ------------------------------------- | -------------------------------------------------------------------------- | | **S — Pilot** | Single-zone, shadow-mode capable | Departmental estate, one or two source tools | | **M — Production** | Three failure domains, full HA | Enterprise estate, several thousand CIs | | **L — Bank-scale** | Three-zone, tier-segregated workloads | Full banking estate, strict T1 latency targets, proven at production scale | ## Why we don't publish generic numbers Sizing an event pipeline from averages is how deployments fail at month-end. Dedup-state size varies by an order of magnitude between two estates with identical daily volumes — which is also why we chaos-verify component failover against **your** state size rather than trusting vendor lab figures. Detailed sizing — node counts, per-component resource tables, Kubernetes requests/limits, and GPU sizing where applicable — is produced for your environment during scoping and delivered as part of the customer documentation set.