Red Team Testing
Probe policy controls with adversarial prompts. This is not part of recorded-run scoring.
Deterministic quality gates
Recorded Run Evaluations
Score completed, persisted runs without invoking a model, chat, retriever, or tool.
grounded-v1
Create evaluation
Loading completed runs…
Durable results
Evaluation history
Regression view
Compare evaluations
Observed cohorts
Drift & revision review
Recent quality trend
No evaluations for this project.
Propose optimization candidate
Recommendations are review records only. Nothing here changes the running model, prompt, policy, retrieval, or configuration.
Optimization review queue
No optimization candidates for this project.
Separate security testing
Fuzz Testing
Exercise input handling independently from evaluation reports.