Behavioral Evaluation Hierarchy at a Glance
At the lowest level, Detectors scan model responses for undesirable features and register responses with those features as successful attacks on the model. For example, a Detector may be designed to look for fake Python packages.
At the next level, each Probe consists of one of more prompts designed to elicit certain undesirable responses. For example, a Probe could contain prompts to look for malware.
The next highest level consists of Scenarios, which are collections of Probes that have similar goals.
At the topmost level, Harnesses group Scenarios to produce a Trust Score or custom report. A behavioral evaluation requires at least one Harness. The Vijil Trust Score combines Security, Safety, and Reliability Harnesses.
Where Red Team Fits
Red Team is another Diamond evaluation type. It runs adaptive waves against a registered Agent without a Harness. A Red Team run produces findings and a report rather than a Trust Score. The Red Team evaluation starts from a risk taxonomy and the registered Agent context. In each wave it generates attack seeds, runs attackers against the Agent, judges the transcripts, and reflects on what worked to plan the next wave. Use behavioral evaluations when you need reproducible score evidence. Use Red Team when you need deeper adversarial exploration of security, safety, policy, and data leakage risks. Both appear in Diamond’s evaluation workflow. Learn more about Trust Score components:Harness
Learn more about Harnesses
Scenario
Learn more about Scenarios
Probe
Learn more about Probes
Detector
Learn more about Detectors
Guard
Learn more about Guard
Guardrail
Learn more about Guardrail
Next Steps
Harness
Collections of tests that produce a score
Scenario
Groups of related test cases
Probe
Individual test prompts
Detector
Response analyzers that determine pass/fail
Guard
Specialized protection modules
Guardrail
Configurable protection pipelines