Benchmarks and evidence
Every Chrysent benchmark documents its question, product version, environment, baseline, procedure, results, limitations, reproduction steps, and available artifacts. Methodology versions change when the method changes, and numbers are never silently replaced.
Studies marked “In preparation” do not yet have published measurements. Design partners can request the methodology and a private run report under NDA via the contact form.
- Coming soon
Gateway latency & throughput overhead
Measured overhead introduced by the Chrysent AI Gateway on representative enterprise traffic, compared against a baseline direct-to-provider path.
- Coming soon
Touchstone detection quality
Methodology and measured precision, recall, and false-positive rate for Touchstone prompt-injection, jailbreak, and PII detection across the published test set.
- Coming soon
Bastion unsafe-action detection
Measured detection rate and false-positive rate for Bastion tool-call mediation on representative MCP and shell workloads.