Those raw reading differences are nowhere close the the maybe 50000 count difference I calculated that you might expect. The first few unloaded readings differ from the loaded readings by about 300, and then the last few unloaded readings are about 4000 higher. Maybe there's a bad connection delivering garbage into the HX711, and then it's all GIGO downstream from there.
A good multimeter could help diagnose whether the fault lay in your connections or your gauge.
The cell works by bending the crossbar between the outer ring and inner ring, and measuring how much the top surface of the bar stretches and contracts. It is a cheap version of a Z-bar load cell.
If the bar or inner ring is sitting directly on a surface, it won't bend, won't stretch, and won't change resistance much with load.