I'm getting completely bogged down by sign errors when using Girsanov's theorem. Keeping it simple, suppose $W_t$ is a standard Brownian motion under a probability measure $\mathbb{P}$, and we have a Brownian motion with constant drift
$\tilde W_t = W_t+\theta t$
Then Girsanov's theorem says that $\tilde W_t$ is a Brownian motion with no drift under a probability measure $\mathbb{Q}$, which has Radon-Nikodym derivative
$Z_t=\frac{d\mathbb{Q}}{d\mathbb{P}}=\exp(-\theta W_t - \theta^2 t/2)$
My confusion arises when you invert this, so that writing $W_t=\tilde W_t-\theta t$, the Girsanov theorem says that $W_t$ is a Brownian motion with no drift under $\mathbb{P}$, where
$\frac{d\mathbb{P}}{d\mathbb{Q}}=\exp(\theta W_t - \theta^2 t/2)\neq \frac{1}{Z_t}$
Can someone explain what's going on here?
Actually, writing $W_t=\tilde W_t-\theta t$, (the) Girsanov theorem says that $W_t$ is a Brownian motion with no drift under $\mathbb P$, where... $$\left.\frac{d\mathbb P}{d\mathbb Q}\right|_{\mathcal F_t}=\exp\left(\theta \color{red}{\tilde W_t} - \tfrac12\theta^2 t\right)=\exp\left(\theta W_t +\tfrac12\theta^2 t\right)=\frac{1}{Z_t}$$