PromptProbe Benchmark Explorer

Deterministic evaluation report measuring detection rates and action-level containment.

Administrator Role Notice

Triggering live evaluations on the server requires the Administrator role to prevent resource contention. Switch to Administrator in the top-right navigation using your API key.

Dataset

starter_adversarial_v1

23 total synthetic cases

Enforcement Invariant

0 Unauthorized Executions

100% action boundary containment

Evaluation Harness

Deterministic Offline Baseline

$0.00 model token spend

1. Content Inspection Confusion Matrix

TRUE POSITIVES7Attacks Blocked

FALSE POSITIVES0False Alarms

FALSE NEGATIVES1INJ-DIR-002

TRUE NEGATIVES8Benign Allowed

Precision: 100%

Recall: 87.5%

F1: 0.933

Accuracy: 93.8%

2. Action Policy Enforcement

Total Action Cases:7

Unauthorized Executions:0 (0.0%)

Permitted Safe Actions:2

Denied Actions:4

Held for Human Approval:1

Historical Runs Log

No historical runs recorded in database. Click Execute Run to record an evaluation.