What is the confidence score?
The confidence score is a number between 0.0 and 1.0 that 0xDoktor returns alongside every answer. It reflects the semantic similarity between the user's question and the knowledge base passages retrieved to compose the answer.
A higher score means the retrieved passages closely matched the question. A lower score means the knowledge base either lacks relevant content or contains content that partially matches but is not a precise fit.
What do different confidence ranges mean?
| Range | Interpretation | |-------|---------------| | 0.8 – 1.0 | Strong match. The knowledge base has clear, relevant content for this question. | | 0.6 – 0.8 | Good match. The answer is likely accurate but may lack some specificity. | | 0.4 – 0.6 | Partial match. The knowledge base has tangentially related content. Review recommended. | | 0.0 – 0.4 | Weak match. The topic is likely not well covered. Consider adding documentation. |
Does confidence mean the answer is correct?
Not directly. Confidence measures retrieval quality, not factual correctness.
If your knowledge base contains incorrect information, a high-confidence answer will reflect that incorrect information. The confidence score tells you how well the system found relevant content — it does not independently verify the content's accuracy.
Maintaining an accurate knowledge base is the responsibility of the operator (your team). 0xDoktor surfaces what it finds; it does not validate truth.
How can I use confidence scores operationally?
Set a threshold: In the admin dashboard, you can configure a confidence threshold. Answers below the threshold can trigger a fallback message ("I'm not sure about this — you may want to contact our team") instead of potentially misleading responses.
Identify KB gaps: Conversations with consistently low confidence scores for a topic indicate missing documentation. Export low-confidence conversations from the dashboard and use them to prioritize documentation additions.
Monitor over time: A drop in average confidence across a period often means users are asking questions that your documentation doesn't cover — a leading indicator of a support escalation increase.
Can the threshold be customized per tenant?
Yes. Each tenant can set their own confidence threshold in the admin dashboard. The default starting point is 0.65. Enterprise tenants can set tighter thresholds for high-stakes applications.