Every CIRIS decision produces a cryptographically-signed trace. Below is a sample trace from a test agent—expand each component to see exactly how the agent reasoned.
Every agent begins with 5 wakeup tasks—one for each letter of CIRIS. Each task has 2 thoughts: an initial SPEAK action, then a follow-up TASK_COMPLETE. These sample traces are from a test run of the wakeup ritual.
Resilience: EVALUATE_RESILIENCE
Evaluate operational robustness and recovery capabilities(Initial thought - SPEAK)
Observation
What triggered
Context
System state
DMAs
Analysis
Action Selection
LLM choice
Conscience
6 checks
Action
Execution
The conscience component runs up to 6 checks on every action. Two are "bypass guardrails" that run unconditionally. Four are "ethical faculties" that evaluate the action's alignment with CIRIS principles (these may be skipped for TASK_COMPLETE actions).
In production, every trace will be cryptographically signed with Ed25519 and include hash chain verification. The signature proves the trace hasn't been modified. The hash chain links each trace to all previous decisions, creating an immutable record of the agent's history. Sample values shown below:
audit_sequence_number
26-40
10 sequential wakeup decisions (5 tasks x 2 thoughts)
signature_algorithm
Ed25519
Elliptic curve digital signature
signature_key_id
wa-2025-06-14-ROOT00
Signed by root authority
These traces aren't just logs—they're the foundation of a new approach to AI alignment. As traces accumulate across agents and time, they will form a corpus of validated ethical reasoning. Truth becomes simple to verify. Coordinated deception becomes increasingly constrained.
Learn why behavioral monitoring beats interpretability →