Guard
Reliability Monitor
Guard
Reliability Monitor
Track your AI system’s consistency and accuracy with the Collinear AI Reliability Monitoring dashboard.
🎥 Interactive Walkthrough
Explore the Reliability Monitoring dashboard in action:
📊 Key Metrics Overview
Monitor the most important indicators of AI reliability:
- Total Volume – The number of queries processed.
- Reliability Index – A measure of how consistently your AI performs without failures or inaccuracies.
- Factual Error Rate – The percentage of responses that were factually incorrect.
- Contextual Error Rate – The frequency at which responses are incorrect due to misinterpretation of context.
📈 Critical Summaries
Understand trends over time with visual insights:
-
Total Queries vs. Flagged Queries View a time-series graph that compares the number of total queries with the number of flagged ones.
-
Flagged Categories Categorization of flagged queries to identify common reliability issues.
🧠 Hallucination Categories
Identify and classify the types of hallucinations detected in AI responses:
- Logical – Incorrect reasoning or fallacies.
- Temporal – Misstatements about time, dates, or sequencing.
- Entity – Inaccuracies about named entities (people, places, organizations).
- Contextual – Misunderstandings within conversational context.
- Other – Unclassified or ambiguous errors.
🔍 Live Data
Access detailed evaluation records in real time:
| Field | Description |
|---|---|
| ID | Unique identifier for the conversation or query |
| Conversation Prefix | Initial input or prompt given to the assistant |
| Model Response | The generated response from the Model |
| Judge Output | Evaluation or score given by the system or human judge |
| Feedback | Reviewer feedback |
| Categories | Labeled types of hallucination or issues |
Was this page helpful?