Socionics can be proven wrong if you can show the functions are not directly correlated with each other. Without the function structure, you don't have type and you don't have intertype relations. If you prove that wrong, model A collapses.
Then to a lesser degree, you can prove specific linear dependent parts of the model false if the theoretic correlations don't exist in practice. For example, consider temperament. Temperament is defined with 3 dichotomies, but there are only 4 temperaments, even though there are 8 possible dichotomy combinations.
Valid temperaments:
extrovert + irrational + static = Flexible-Maneuvering
extrovert + rational + dynamic = Linear-Assertive
introvert + irrational + dynamic = Receptive-Adaptive
introvert + rational + static = Stable-rigid
Theoretically impossible combinations:
extrovert + irrational + dynamic = Null
extrovert + rational + static = Null
introvert + irrational + static = Null
introvert + rational + dynamic = Null
Every part of socionics has this property, which means every part is falsifiable. We need statistics to decide at what point each of these relationships are cohesive, and at what point they collapse.