This is a surprisingly hard question to answer that @TheSpaceGal has asked. Reentry predictions are very complicated, which is why predictions differ and why the range of reentry orbits has been so broad. Some of the primary factors: (1/n) https://twitter.com/TheSpaceGal/status/1391149315830149121

The model itself. There are many different models that try to mathematically simulate the atmosphere. Each model has different characteristics, strengths, weaknesses, and oddities. Then the model has to be given inputs and assumptions such as...

What is the sun doing (aka space weather) the sun's output varies slightly and that affects the atmosphere making it slightly bigger or smaller thereby changing its density at the satellite's altitude, which in turn affects the drag.

The shape of the satellite (or rocket stage in this case). Pointy shapes have less drag an will take longer to reenter. Flat shapes have more drag and will reenter sooner (every kid has tried experimenting with their hand out of a car window);

the rocket stage is tumbling, so we don't fully know whether its skinny tip is toward the "wind" or whether it is sideways with its big fat side toward the "wind".

The mass of the rocket stage. This is a spent rocket stage and we don't exactly know what is mass is, how much leftover propellant is onboard, etc. Lighter objects will get slowed down more by the atmosphere.

... and speaking of propellant, if any leftover propellant is leaking, it would serve as a mini-rocket pushing the stage around, slowing it down or speeding it up.

Even the surface finish of the stage could have some minor impacts on the drag the stage produces.

Many of these effects are cumulative and small changes in one can impact the other. Differences also impact how gravity affects the rocket stage since the earth's gravity is not perfectly uniform.

So in short, the surprising thing not that the models are different, but that that we can make educated enough assumptions that the models agree so closely. (end)