When uploading an experiment dataset to Galileo, each row must include the question, the model's answer, and the ground truth answer; those three fields are what the scorers compare. Context is optional but worth including whenever retrieval feeds the answer, since context-adherence scorers need it. Validate the JSON schema before upload: a misnamed field does not error loudly, it just produces empty scorer results. Keep rows self-contained; cross-row references break when Galileo treats each row as its own trace.

Context: Official docs (rungalileo/sdk-examples, upload experiment README): documents the upload schema that fails silently when wrong: the dataset JSON needs question, model answer, and ground truth fields, with context optional.