What are scores and metrics and when should I use each?
Last updated: April 13, 2025
Scores vs Metrics
When logging evaluation data in your pipeline, you have two options:
Type | Use Case | Value Range |
Scores | For normalized quality measurements that will be visualized and analyzed across runs | Must be between 0 and 1 |
Metrics | For raw measurements like counts, lengths, or unbounded values | Any numeric value |
How to Log Metrics
You can log metrics using the span.log() method with the metrics parameter:
span.log(metrics={
"citation_count": 5,
"response_length": 256
})Note: While metrics can be logged with any numeric value, aggregation and visualization of custom metrics is not currently supported in the platform.
How to Log Scores
Scores should be used when you want to track normalized quality measurements that can be compared across runs. Always normalize scores to be between 0 and 1 before logging:
span.log(scores={
"relevance": 0.85,
"accuracy": 0.92
})