uploaded predictor persists in cloned analysis bundle even if removed [bug report]

[jsmentch]

(using the web interface)

I am seeing some odd behavior where if I clone an analysis with a custom predictor that I uploaded, this feature will persist in the cloned analyses even if it is unselected, like a "ghost predictor". For example:

• as a test: rq2o2 was cloned from another analysis, say b758b. I then removed all predictors and only added one 'face' predictor and submitted this analysis. You can see that 'speech_edited' custom feature is still listed as a predictor in this new analysis that should just have 'face'. Interestingly, 'speech_edited' is not included in the bids stats model but it is downloaded in the bundle.

• another example is npqie. I ran an older analysis with a 'speech_edited' custom predictor, then I re-uploaded an updated 'speech_edited' custom predictor and then cloned the older analysis to create npqie and tweaked it to how I wanted it. You can see though that 'speech_edited' is listed twice under the list of features, but only the new version is downloaded in the bundle and only one is in the BIDS stats model.

may be specific to my unique case of uploading a feature, creating an analysis, re-uploading a feature of the same name, then cloning the analysis that has this out of date feature?