CAIN-42 CAIN StudioCAIU · Module 2 of 5 · 16 min
Reading an evidence-derived fleet risk score (0–1000)
CAIN's ACTUARY KEYSTONE scores a fleet from its real decision records: what agents tried, what CAIN allowed, held and blocked, and how volatile behaviour is. Every factor is a documented rule in the code.
High-risk paths are actions that can move money, run code, delete, change authority or send data out (transfer, exec, delete, grant, export, send_email…). Read-only paths (read, list, search, query) are treated differently.
A score is only meaningful with its inputs: the period covered, the number of decisions, and the share of high-risk actions that were allowed without a gate.
Key points
- Score derived only from recorded decisions.
- High-risk = money, code, delete, authority, exfiltration.
- Always read the score with its window and volume.
- Every factor is a documented rule.
In the CAIN-42 codebase
keystone/actuary.py— HIGH_RISK / READ_ONLY patterns and the scoring rules
Practice check
Practice 1 · choose one
Which action is high-risk under ACTUARY's rules?
Practice 2 · true or false