Best Practices
This section covers how to get the most out of your Nyckel prediction endpoints — whether you are working primarily in the UI or integrating via API.
In this section
Get reliable predictions
Label design, data quality, and what makes a well-defined classification task.
Improve accuracy over time
How to use samples, annotations, and feedback to increase accuracy.
Confidence scores and thresholds
Understand what confidence means and how to set thresholds for your use case.
The core idea
Nyckel is built for scoped classification tasks with measurable confidence. Unlike open-ended AI tools, every prediction comes with a confidence score you can act on. You can set thresholds, route low-confidence predictions to human review, and improve the endpoint over time with feedback.
The more specific and consistent your labels, the more reliable your predictions will be.