Data preparation issues

Start with the invariant named by the exception

Do not reshape, truncate, reorder, or fill scientific arrays merely to make an export complete. Check these identities directly:

• every embedding and vector field has one row per observation;

• expression rows follow the same observation order;

• expression columns follow the declared gene identifiers;

• connectivity is square and matches the observation count;

• categorical and continuous metadata use their declared types;

• dataset and gene identifiers are explicit and stable.

Export completes but the viewer rejects the folder

Confirm that the selected directory is the export root and contains dataset_identity.json and obs_manifest.json. Then inspect the first failed request in the browser Network panel; its exact filename and HTTP status distinguish a missing artifact from malformed content.

Values look scientifically wrong

Compare a small set of observation and gene identifiers before and after export. Treat transposition, silent reordering, category-code mismatch, non-finite coordinates, and unexpected quantization as data-contract failures.

See Input requirements (global), Output format specification (exports directory), and Troubleshooting (prepare/export).