Proof, not promises.

Methodology & Evidence

Every performance and security claim on this site carries a label showing where it comes from - and the numbers behind it are published here.

How to Read Our Claims

Four labels. Every claim on this site carries one of them.

MEASURED IN LAB

Result of a repeatable test on our reference systems. Method and rig disclosed.

ENGINEERING TARGET

A design goal we optimize for. Not yet a published measurement.

SELF-ASSESSED

Our own evaluation against a published framework, with full scoring disclosed.

THIRD-PARTY VERIFIED

Confirmed by an external organization. Scope of the verification stated.

Measured in Our Lab

Repeatable tests on reference systems and on our own production fleet. The numbers below are counter and wall-clock readings, not projections.

Agent Footprint

Idle CPU0.02%
Resident RAM~24 MB
CPU during backups21-28%
Disk throughputIdentical with and without the agent

CPU during backups measured while saturating a 1 Gbps link - the network was the bottleneck, not the agent. Rigs: AMD Ryzen 7 7700X with Kingston NVMe (Windows 11) and AMD Ryzen 9 9950X3D (Debian 13).

Recovery Speed

Instant iSCSI Mount (Linux)5-20 s
Instant iSCSI Mount (Windows)5-120 s
Proxmox Instant BootMount <=15 s + ~120 s OS boot
BMR to login screen (Linux)4-6 min
BMR to login screen (Windows)7-9 min

Mount times independent of volume size (tested 30 GB to 1 PB). BMR measured wall-clock from recovery ISO to login screen on 2 TB reference disks, five runs each (Linux: 5, 6, 5, 6, 4 min - Windows: 7, 8, 7, 9, 8 min). The operator typed the recovery key manually during these runs.

Scale & Deduplication

Largest protected volume1 PB (Ceph cluster)
Devices in a single realm24
Physical after dedup (realm key)13.10 TB -> 210 GB
Physical after dedup (per-device key)8.15 TB -> 207 GB
Physical after dedup (realm key)30.15 TB -> 620 GB
Physical after dedup (realm key)3.15 TB -> 177 GB
Incremental backup~2 min at ~5% daily change
Reference VM cadenceBacked up every 4 hours

Physical storage after deduplication measured on production realms. Deduplication ratios are workload-dependent; these are our production numbers, not a promise for a specific workload.

Fleet Reliability

Machines under continuous protection24
Observation period6 months
30-day backup success rate100%
Agent OTA updates delivered~40
Longest uninterrupted run6 months

Our own production fleet, dogfooding against the production control plane. OTA updates counted since the first production deploys in April 2026.

Active Defense

Entropy, simulated ransomware7.51 / 8 (average)
Entropy, ordinary data3.73 / 8
Alert email dispatched5 s after detection
CBT-armed detection trigger~500 MB of encrypted writes

On detection the affected backup is automatically quarantined and marked. Optional network lockdown isolated the device while the management channel stayed alive, and the lockdown persisted across a hard reset until reversed by an MFA-confirmed unlock. Results reproduced on live production systems, including detection with block change tracking armed and the forced-poweroff response.

SELF-ASSESSED

SEAL-4 Sovereign Infrastructure

StateWarden achieves a SEAL-4 rating in our self-assessment against the EU Cloud Sovereignty Framework v1.2.1, with the per-objective scoring published in full. Self-assessment, not an audited certification.

Read the SEAL Sovereignty Whitepaper
SEALLevel 4(self-assessed)
THIRD-PARTY VERIFIED

ETSI Full Membership

Cyberprime has been an ETSI Full Member since 22 September 2026, and the certificate is published. An honest note: ETSI membership signals standards participation - it is not a product certification.

View Certificate

Engineering Targets

What is not yet a published measurement. We optimize for these; we do not present them as results.

  • Recovery-time guarantees under a contractual SLA
  • Deduplication ratios for any specific customer workload (workload-dependent)
  • Agent overhead on links faster than 1 Gbps
  • Write-path overhead of the change-tracking driver
  • Crypto-layer throughput figures

When we have not measured it, we say so - and the list above is the whole of it.

Test Environment

Where the numbers above come from.

  • On-premises lab: Proxmox VE 9 and VMware ESXi/vCenter 8.0.3 hypervisors, with Debian 13 and Windows 10/11 clients.
  • Our own production fleet: 24 machines running against the production control plane.
  • Verification method: repeated runs, results cross-checked against system journals and database records - never accepted from a UI confirmation alone.

Privacy by Design

This website uses only first-party cookies strictly necessary for its operation. Zero third-party trackers, zero external analytics platforms.

Read our Privacy Policy for details.