Passed with one manual review branch.
Correctness and policy compliance stayed above threshold.
Reviewed by Prometa QA · run at 2026-03-15T15:00:00Z
Testing lab
Validate agents with scenario, regression, approval, and benchmark testing before release.
Testing Lab combines manual test runs, golden dataset checks, review scoring, execution telemetry, and release readiness checks in one bounded environment.
3Test cases
1Passed
1Failed executions
2Release-ready agents
Escalation triggered correctly, relevance score dipped on routing note.
Review copy needs refinement before publish.
Reviewed by Claims Reviewer · run at 2026-03-15T16:30:00Z
Variance exceeded target for one peak cluster.
Regression breach blocks release readiness.
Reviewed by Planning QA · run at 2026-03-15T18:00:00Z
Evaluation engine
Exact match, semantic similarity, rule validation, and human scoring can run on the same test asset.
Scoring functions
Exact match · semantic similarity · rule-based validation · human review scoring.
Regression testing
Compare version vs version to prevent silent behavior drift before release.
A/B testing
Run agent_v1 vs agent_v2 or workflow branch vs branch under the same dataset.
Batch runner
Asynchronous evaluation runner supports multiple test cases, versions, and score summaries.
Release readiness
Deployment readiness stays explicit before publish to pilot or production.
Tests passed: yes
Integrations healthy: yes
Governance approved: yes
Rollback path: available
Tests passed: yes
Integrations healthy: yes
Governance approved: yes
Rollback path: available
Tests passed: no
Integrations healthy: yes
Governance approved: no
Rollback path: missing
Execution trace
Trace-backed evaluation view for regression and approval testing.
Trace `/trace-001` exposes orchestration steps, tool calls, decision points, latency, token cost, and suggested remediation without opening backend internals.
Scenario ingestsuccessScenario pack normalized
Forecast API
Peak cluster anomaly detected
1240 in · 318 out · $2.10
model 420 ms · tool 980 ms · orchestration 210 ms
Planner review routingwaiting_reviewEscalated to planner queue
Planner Review UI
Confidence below threshold
220 in · 91 out · $0.60
model 180 ms · tool 220 ms · orchestration 140 ms
Decision path stayed inside registered tools, applied runtime guardrails, and routed to review whenever confidence or policy state required intervention.
Planner scenario payload missing one required pricing field. Suggested fix: Add schema validation in input handler and inject default pricing floor before review.