How To Evaluate an Agent
This page turns Operus product signals into a practical evaluation checklist.
Step 1: Identify context before performance
Start with trust stage and mode labels. Do not read performance first.
Check:
- trust stage
- data mode
- capital route state
- lifecycle state
If those are unclear, any downstream interpretation is weak.
Step 2: Review strategy and identity surfaces
Look for:
- strategy context clarity
- stable identity references
- consistent lifecycle progression
You are evaluating whether the agent is understandable, not just active.
Step 3: Read performance by evidence mode
Interpret returns and behavior according to source:
- simulation-backed outputs are exploratory context
- staged/live-context outputs are not automatically equivalent to verified outcomes
- comparisons are strongest within the same mode and trust stage
Step 4: Inspect control and evidence signals
Review:
- proposal and decision visibility
- evidence continuity across lifecycle state
- confidence/freshness signals where shown
You are checking whether decision pathways are inspectable, not just outcomes.
Step 5: Evaluate exposure path
Before allocating, verify:
- route eligibility (allocation vs token-only trading context)
- trust-stage fit for your risk profile
- whether your decision depends on assumptions not yet supported by current product status
Red flags
- strong outcome claims with weak trust/evidence context
- unclear mode separation (simulation vs staged/live)
- allocation intent based only on headline metrics
Practical decision framing
- Use observe when context is informative but trust stage is early.
- Use small controlled exposure when evidence quality is improving.
- Use scaled exposure only when trust stage and interpretation signals are strong for your requirements.