AmbitIndex Methodology
Last updated: November 17, 2026
This page explains how AmbitIndex collects, summarizes, and presents cloud compute telemetry. The measurements describe observations from specific test instances operated by AmbitIndex. They are not rankings and should not be interpreted as measurements of an entire provider’s infrastructure.
1. Test scope and configurations
AmbitIndex currently collects measurements from one basic-tier test instance at each of the following providers:
| Provider | Test configuration |
|---|---|
| Vultr | 1 vCPU, 1 GB RAM, 25 GB storage |
| DigitalOcean | 1 vCPU, 1 GB RAM, 25 GB storage |
| Linode | 1 CPU core, 1 GB RAM, 25 GB storage |
These configurations provide context for the observations. They do not establish that every provider plan, region, host, or customer workload will produce similar results.
2. Collection schedule
Each test instance is scheduled to submit telemetry approximately every three hours. Collection is designed to be low impact and uses a small number of requests. A missed or unsuccessful submission can result in a longer interval between observations.
3. HTTPS response measurements
During each collection run, the test instance makes seven sequential HTTPS requests to the AmbitIndex measurement endpoint. Each request is timed from immediately before the request begins until the HTTP response headers and status are received. The response body is not fully downloaded as part of this timing.
These are application-level HTTPS response timings observed by the client. They include connection and server-response overhead, so they are not measurements of network latency alone and are not full webpage or response-body download times.
HTTPS response maximum
The longest of the seven request timings in a collection run. This is a maximum, not a 95th-percentile (p95) measurement.
HTTPS response variation (jitter)
The mean absolute difference between consecutive request timings after the seven timings are sorted from lowest to highest. It describes variation among the measured HTTPS response durations. It is not the packet interarrival jitter defined by RFC 3550.
HTTPS response spread
The difference between the highest and lowest of the seven request timings in a collection run.
4. CPU steal
CPU steal is calculated from changes in the cumulative CPU time counters exposed to the guest operating system through /proc/stat. It estimates the share of elapsed CPU time during which the virtual machine was ready to run but the hypervisor scheduled other work.
The reported value is calculated over the interval between successful telemetry baselines:
CPU steal (%) = 100 × steal-time delta ÷ total CPU-time delta
This is an interval average, normally covering approximately three hours, rather than a one-second snapshot. The baseline is updated after a successful telemetry submission. After a reboot, counter reset, or initial run, a new baseline is established and no interval value is reported until a subsequent valid measurement is available. CPU steal is guest-visible telemetry and does not reveal the complete state of a provider’s physical host.
5. AmbitIndex Alpha eligibility and aggregation
The current Alpha aggregation method is ambitindex-alpha-v2-14d. It evaluates observations from a rolling 14-day window.
- Only the HTTPS response maximum metric is used to determine Alpha eligibility.
- A provider must have at least 30 eligible observations in the 14-day window.
- At least one distinct reporting instance is required.
- The displayed aggregate is the median of the eligible HTTPS response maximum observations.
Jitter, response spread, and CPU steal provide additional context but do not determine Alpha eligibility. Sample count and distinct instance count are shown so readers can see the amount and breadth of the available reporting.
6. Historical charts and Alpha are different views
Historical charts show daily UTC medians over a period of up to 90 days. Alpha eligibility uses a separate rolling 14-day window and its own minimum sample requirement. A historical chart may therefore contain observations that are outside the current Alpha window. Historical chart coverage and Alpha sample counts should not be treated as interchangeable.
7. How to interpret the measurements
AmbitIndex reports measurements from its own test instances under the conditions observed during collection. Results can be affected by instance configuration, location, software, network paths, neighboring workloads, provider infrastructure, and conditions at the time of measurement.
Repeated observations from one instance are useful for tracking that instance over time, but they are not independent measurements from multiple physical machines. A distinct instance count indicates the number of reporting instances represented; it does not establish the number of physical hosts involved.
These measurements are not a guarantee of performance, a substitute for testing an application under its own workload, or a statement about every customer or location served by a provider.
8. Transparency and changes
AmbitIndex publishes measurements to let readers examine the observations and interpret them for themselves. It does not convert the telemetry into a provider ranking or declare a universally superior provider.
Methodology changes that materially affect collection, metric definitions, eligibility, or aggregation are documented so that readers can understand how published results are produced. Historical records may remain available when a collection method changes; data from different methods or measurement periods should not automatically be assumed to be directly comparable.