Hetzner cpx22

• Single deployment; not an independent three-instance cohort.
• No hardware attestation.
• Pilot memory implements read/write; copy workload deferred.
• Network endpoints must be supplied by the operator; no public third-party servers are used.
• Database latency distribution is not available from this pilot parser; average latency only.
• This profile is not a full implementation of all proposed methodology tests.
• Resource-constrained environment: CPU workers and RAM reflect affinity/cgroup limits, not host capacity. Private resource-limits evidence retains both; do not merge container/VM or architecture cohorts.
• Short profile: does not establish 24-hour consistency.
• This report received AI-assisted evidence review under user-delegated authorization. This is not independent human verification or hardware attestation.
• This pilot measures its recorded test window. It does not establish 24-hour consistency, long-term reliability or provider-wide performance.
• Network tests use two controlled endpoints, one stream, a 100 Mbit/s rate setting and a requested 64 MiB byte target per direction. Reverse-mode polling can exceed that byte target; actual sent and received bytes govern traffic accounting. These results do not measure maximum link capacity.
• Provider, region and SKU were matched to retained deployment records. Missing detected values and deployment locations within a detected Hetzner network zone use operator-supplied context and remain labelled supplied. The original detected network zone is retained privately; the detected datacenter zone remains in the report. Control-plane records and checksums do not attest the physical host or establish physical-host independence.
• Package versions and available guard and journal evidence were reviewed. These records do not establish continuous absence of package activity throughout the test window.
• The pilot includes memory read/write and average PostgreSQL latency. Memory-copy tests and PostgreSQL latency percentiles are not implemented. Skipped consistency and missing measurements are not scores.
• These network observations were collected during provider-reported maintenance on core routers. They are not a normal-path baseline. The status notice does not establish that maintenance caused any individual measurement or any CPU, memory, disk or database result.

Original full benchmark deployment · 17 Oct 2026. These controlled iperf path checks used a 100 Mbit/s rate setting and do not measure internet speed or maximum link capacity. RTT is the probe mean; jitter and packet loss were not measured. The original report JSON retains these measurements.