Skip to main content
Applies to:
  • Plan -
  • Deployment -

​
Summary

Issue: An unexpected score column appears in Logs after an online scorer automation runs, or remains visible after an automation or scorer was deleted. Cause: When an online scorer runs, its result is written to the log row under scores.<name>. That persisted score key comes from the scorer function name or the scorer’s returned score name, not from the rule name unless those names happen to match. Self-hosted deployments running data plane versions earlier than v2.13.1 can also show empty columns named after configured online scoring rules. Resolution: For self-hosted deployments, upgrade to data plane v2.13.1 or later so configured rule names do not create empty columns. To change the score keys written to future log rows, rename the scorer function or return explicit { name, score } objects. Existing log rows are not rewritten, so historical score columns may remain visible while those rows exist.

​
How score columns are named

When an online scorer runs, Braintrust writes the scorer result into the log row’s scores object. If the scorer returns a single value (a number, true/false, or null):
Braintrust uses the scorer function’s name as the score key. It does not use the online scoring rule name unless the scorer function happens to have the same name. The same applies when the scorer returns a score object without a name:
Braintrust uses the scorer function’s name as the score key. If the scorer returns a score with an explicit name:
Braintrust uses that name field as the score key. If the scorer returns multiple named scores:
Braintrust writes one score key for each returned name. The scorer function’s name is not used here — only the returned name values become column keys.

​
What to check

Use these checks to track down the source of the unexpected column.
  • Confirm whether the column appears on new logs, only on historical logs, or both.
  • Review active rules in Settings > Automations.
  • Compare the unexpected column name against the scorer function name and any names returned by the scorer.
  • If an empty column matches a rule’s name on a self-hosted deployment running a data plane version earlier than v2.13.1, upgrade to v2.13.1 or later.
  • If the unwanted column contains actual score values, rename the scorer function or change the names it returns.
  • Refresh Logs and verify the behavior against a newly created log row.

​
Notes

  • Deleting an automation stops future runs. It does not remove score values already written to past log rows. The same is true if you rename or delete a scorer: historical rows are not rewritten.
  • A column can stay visible as long as the rows currently in view still contain that score in their data.
  • If you’ve changed a rule or scorer and the Logs table looks out of date, a hard refresh in your browser forces the UI to reload the latest configuration.