Baselines

Calibration by workload class

X-40™ governance is baseline-driven. Baselines are calibrated per workload class (facts, math, unknowns, attack, or your domain-specific prompt classes).

Baselines define “normal” behavior for a class of prompts or inference runs. X-40™ measures drift relative to baseline and enforces policy thresholds.

Onboarding typically requires:

LLMs: 10–20 prompts per class
ML: representative logs or a sample telemetry file