I acquire the raw JP2 imagery files from the Sentinel-2 project. Let's say I've selected a tile for a given sampling period and band (e.g. B01). Then, after using Python tools provided from here, I compare my JP2 files to the TIFF files (corresponding to the exact same spatial area corresponding to the JP2 file) provided via get calls to the Sentinel Hub AWS S3 source. These two files - regardless of sampling period, location, or (usually) band are different in the following ways: 1] array dimensions (e.g. JP2=1830x1830 vs. TIFF=10980x10980) and 2] measurement values (e.g. JP2=[[1000,1001,1023,...]] vs. TIFF=[[0.780,0.888,0.901]]).
This means two things. First, some kind of interpolation is going on from the Sentinel-2 raw data format to resample a coarser array to one comparable to more higher resolutions bands like B02, B03, etc. Also, the array value themselves corresponding to a given TIFF image have also undergone some form of normalization/standardization/regularization/equalization.
What does Sentinel Hub do - or perhaps the Python Sentinel-2 toolkit - to the images to make this large difference happen compared to the supposed raw JP2 format? That is, why the radical difference between the JP2 and TIFF files?