Let $(M , g)$ a $n$-dimensional pseudo-riemannian manifold. As $g(p)$ is a bilinear mapping from $T_pM \times T_pM$ to $\mathbb{R}$, we can get a basis $$ {\left\{{\left(\frac{\partial}{\partial x_i}\right)}_p\right\}}_{i = 1}^n $$ on $T_pM$ such that $$ g(p)\left({\left(\frac{\partial}{\partial x_i}\right)}_p , {\left(\frac{\partial}{\partial x_j}\right)}_p\right) = 0 $$ for each $i , j \in \{1 , \ldots , n\}$ such that $i \neq j$. It allows us to define the index (or signature) of $g(p)$ because it is independent of the choice of basis in $T_pM$, a fact known classically as Sylvester's law of inertia.
Well, many texts talk about index of $g$ strightly, and define for instance riemannian metrics as pseudo-riemannian metrics which index coincide with the dimension of the manifold (or signature ($n , 0$) in this case).
How can I show that it is independent of the point $p$ in $M$? As it can be observed, I have not used neither $g(p)$ is non-degenerate or symetric nor the fact $T_pM \cong {\mathbb{R}}^n \cong T_qM$, as isomorphism of vector spaces ($p , q \in M$), for defining signature of $g(p)$ (for $p \in M$ fixed). Must I use one of these statements to show that the index of $g(p)$ and $g(q)$ coincide for each $p , q \in M$?