About this report: All metrics were produced by the
Cloud World Model
simulation engine — a deterministic, capacity-aware model that calculates
CPU utilisation, P50/P95 latency, throughput, error rates, and cost from first-principles
resource profiles without provisioning any real cloud infrastructure.
Official hourly rates from AWS, GCP, Azure, OCI, and DigitalOcean pricing calculators
(June 2026, us-east-1 / us-central1 / East US / us-ashburn-1 / nyc1) are embedded
directly in the engine and validated weekly.
See
cloudworldmodel.ai/fidelity
for the rate-check basis: the ±10% tolerance applies to modeled hourly rates for standard
compute and database SKUs against published provider on-demand list prices. Always-Free and
request-billed SKUs are excluded. This is a unit-rate check, not a whole-workload or actual-bill
guarantee; usage-dependent storage, network charges, negotiated prices, and discounts are not
covered by this tolerance.
For the separate company-owned real-AWS cwm-bench comparison, see
Accuracy Benchmark.
Mid-tier compute and burst headroom: The GCP and Azure configurations use
e2-standard-2 and Standard_D2s_v3 respectively — 2 vCPU shapes chosen to reflect
realistic mid-tier deployments (see the
Capacity column in the scenario table above).
Smaller per-instance headroom means these clusters reach higher CPU utilisation at
2× burst traffic, which is why their P95 latency (227 ms / 184 ms) and error
rates (10–12%) are elevated compared to the larger m5.large and E4.Flex shapes.
This is expected behaviour, not a measurement artefact: the numbers reflect the trade-off
between instance cost and burst tolerance inherent to mid-tier compute.