The argument from stochasticity conflates classificatory underdetermination with theoretical constitution: multiple models can track the same real phenomenon without that phenomenon being model-dependent.
?Rate how convincing each reason is below to see the overall strength.