Risk scoring
A risk score on
every single Reaction.
Per-Action governance tells you what a write did. Quadrazene scores the whole run. A composite 0-100 from six weighted factors, so compliance can find the 25 runs worth investigating without scrolling past hundreds of green ones. Cross a threshold and the run routes itself to a human.
Six factors, weighted and summed
What goes into the score.
Each factor contributes a bounded amount. The sum lands the Reaction in a band. Weights are data, not code.
Classification
0-30Data sensitivity of the System touched: public → internal → confidential → restricted.
Side effects
0-30What the atoms actually do. Writes, payments, outbound comms, and their risk class.
Reversibility
0-20Payments and outbound communications dominate. Reads barely register.
Novelty
0-10How rarely this Formula has run versus the tenant median. First-time runs score higher.
Trust events
0-20Filter activity, weighted: block ×8, redact ×4, warn ×1.
Guardrail bypass
0-20Write overrides, skipped HITL, disabled policies. Anything that loosened the rails.
Risk Center
Find the runs worth looking at.
What the score drives
A number that does something.
Auto-HITL routing
Cross the threshold and the runner opens an Approval in the Inbox, with a one-line explanation of which factors pushed it over.
Risk chip everywhere
A colored badge sits on every Reaction in Records and reaction lists. Tooltip shows the per-factor breakdown.
Tenant-tunable weights
Every weight is a setting. A compliance officer can dial classification up or novelty down without an engineer.
Replay on history
Reapply new weights to the last 7 / 30 / 90 days so a policy change reflects on past runs, not just future ones.
Bands, not noise
low / medium / high / critical so dashboards and alerts speak in plain risk language, not raw integers.
Feeds Advisory
High-risk patterns roll into Advisory's prioritization. The platform recommends what to tighten.