Folders named single are refused, but existing lab datasets use them #24

@IvanHCenalmor, raised at Guillaume's request. This is a question for you; nothing was changed here.

What happened

To fit MorphoFeature's model-input scaling from your saved histograms (plane_histograms.csv + run_manifest.json, MorphoFeature decision D70), I ran run_experiment on MorphoRelate's data/hpmec_perfusion/images. That dataset is laid out as HPMEC_ctrl/single/Position_1.tif, i.e. condition / biological repeat named single / image. Preprocessing refused it:

reserved_folder_name: "single" is reserved for missing hierarchy levels and cannot be used as a folder name.
paths: ('HPMEC_cond2/single', 'HPMEC_ctrl/single')

This matches docs/contract.md §1 ("A folder named single (in any upper/lower case) is an error"). Copying the images without the single level (HPMEC_ctrl/Position_1.tif) gives exactly the labels the dataset intends: condition HPMEC_ctrl, repeat single, technical repeat single. MorphoFeature's fit from your table then equals the fit on the images, bit for bit.

The question

A folder named single yields the same label a missing level gets. Lab data written for MorphoRelate, and possibly other existing folders, use that layout, so users would have to restructure their data before preprocessing. Would you consider one of these?

1. Accept a folder named single as an explicit single label. Keep refusing the cases that really are ambiguous, for example a single folder next to sibling repeats, or case variants like Single.
2. Keep the refusal, but improve the message: say that the folder's images can move up one level and will get the same labels.
3. Keep it as is. I'll then document in MorphoFeature's run tool that lab folders must not use single.

You know why the rule exists better than I do. If there's a case where accepting it would mislabel images, option 1 is off the table.