To find the $n^\text{th}$ roots of a complex number, we can first express it in polar form (I'm assuming $r=1$ for brevity; it doesn't matter for my question):
\begin{align} e^{i\theta} &= \cos\theta + i\sin\theta \\ &= \cos\left(\theta + 2\pi k\right) + i\sin\left(\theta + 2\pi k\right), \qquad k\in\mathbb{Z} \end{align}
and then use De Moivre's formula:
\begin{align} \left(e^{i\theta}\right)^\frac{1}{n} &= \cos\left(\frac{\theta + 2\pi k}{n}\right) + i\sin\left(\frac{\theta + 2\pi k}{n}\right), \qquad k\in\mathbb{Z} \end{align}
However, when we express a complex number in polar form we could also shift the $\cos$ and $\sin$ arguments by different multiples of $2\pi$; this is certainly still valid: \begin{align} e^{i\theta} &= \cos\theta + i\sin\theta \\ &= \cos\left(\theta + 2\pi a\right) + i\sin\left(\theta + 2\pi b\right), \qquad a,b\in\mathbb{Z} \end{align}
but it does it not follow that:
\begin{align} \left(e^{i\theta}\right)^\frac{1}{n} &= \cos\left(\frac{\theta + 2\pi a}{n}\right) + i\sin\left(\frac{\theta + 2\pi b}{n}\right), \qquad a,b \in\mathbb{Z} \end{align}
This is obviously false; for example, taking $\theta = 0$, $n=3$, and setting $a=1$, $b=0$, we get $-0.5 + 0i$ as a cube root of unity.
I know that the answer has to be "this is not a correct application of De Moivre's formula", but an explanation of why this is the case would be nice.