Hi, many test records are variants (e.g. a different flavour or size) of products that appear in unlabelled_data.csv. The guardrail document (§4, Variants) says shared copy may be reused “only when it is true for every variant”, while the data page says unlabelled_data.csv should not be treated as a verified answer source for a test record.Is it acceptable to use unlabelled_data.csv to recover a fact for a test record only when that fact is identical across all of its sibling variants (e.g. a certification or packaging type), while never copying variant-specific values such as weight, dimensions, flavour or shade? We want to stay within the spirit of the guardrails. Thank you! @zindi @meganomaly