In a DEG run, count tables carried generic/duplicate labels (e.g. 0, 1, 2, or identical Counts Table names), producing non-unique identifiers that prevented DESeq2 from running. Orbit eventually worked around it by downloading and re-uploading the count tables with unique names, but that's a manual detour.

Orbit should preserve original sample identifiers from the upstream collection and validate that labels are unique (and non-generic) before launching DESeq2, rather than failing and recovering by re-upload.

Related to #208 (collection handling) and to the fold-change interpretation issue #289.