Let $A$ be an abelian variety over $\mathbb{F}_p$. Then of course for every natural number $i$, we have that $\# A(\mathbb{F}_{p^i})$ divides $\# A(\mathbb{F}_{p^{i+1}})$.

But MAGMA says this is false: Here is my code:

P<x> := PolynomialRing((FiniteField(3)));
J := Jacobian(HyperellipticCurve(x^6 - 2 * x^5 + x^4 - 2 * x^3 + 6 * x^2 - 4 * x + 1));
for j in [1..10] do;
    Order(BaseChange(J, FiniteField(3, j)));
end for;

And the result is:

19 57 1444 5529 59299 467856 4976347 43264425 394975876 3458495577

What is wrong?