Methodology
A consistent measure of agent performance, grounded in operational evidence.
Two perspectives.
Evaluate the agent and the organization supporting it separately.
- AgentScore
- A 0–1000 score for each agent’s operational performance.
- ImplementationScore
- A 0–100 measure of the organization’s deployment context, including integration, governance and data foundations.
From records to results.
- Step 1
Connect
Read supported sources, upload a CSV or submit API events.
- Step 2
Validate
Check metric names, dates and units before scoring.
- Step 3
Measure
Apply the versioned model across eight factors.
- Step 4
Inspect
Trace the result to its supporting evidence.
Eight factors.
The asam_v2.1 model uses fixed weights shared across customers. Changes require a versioned release.
- Efficiency
- Quality
- Business impact
- Autonomy
- Reliability
- Durability
- Learning
- Governance
Visible limits.
Minimum evidence
Agents with fewer than 50 recorded events show as collecting data. Scores below 60% confidence are provisional.
No manual overrides
Customers and Orbitex cannot edit a score. Recompute runs the model again.
Retained evidence
Original source records remain available. Remove unnecessary personal data before submitting them.
Explore the API behind the scores.
Discuss your workflow and the evidence you need.
Developer reference