How evidence is verified

Versioned methods, preserved source artifacts and reproducible summaries.

The evidence process

Plan and environment: Exact recorded configuration
Provider, plan, region, architecture and cohort
CPU: Events per second
Pinned sysbench method; single and multi-thread measurements remain separate
Storage: Read and write IOPS
Pinned fio method against a dedicated scratch file
Network: Mbps by direction
Only permitted endpoints and recorded traffic limits; unavailable tests stay unmeasured
Observation window: Actual start and finish times
No claim of continuous uptime monitoring or guaranteed retesting

Accepted measurements: Median and observed range
Each metric shows its own sample count, unit and exact method
Failed rounds: Missing with a reason
Failures remain visible and never become zero measurements
Comparisons: Compatible cohorts only
Protocol, architecture, plan size and measurement methods must agree
Ratings: No composite provider rating
Short runs do not establish support quality, long-term uptime or provider-wide consistency

Source evidence: Download evidence manifest
Sanitized artifacts are checked against their SHA256 hashes before publication
Corrections: New versions preserve earlier records
A corrected run names the prior version and explains the change
Licences: Separate per source artifact
Third-party tools and data retain their own terms; no blanket dataset licence is granted here