I started reading A tour of subriemannian geometry, their geodesics and applications by Montgomery. There's one thing I don't quite understand, which probably boils down to linear algebra, but I'm just not seeing it.
Let $(Q, \mathcal{H}, \langle \cdot,\cdot\rangle)$ be a sub-Riemannian manifold. A cometric on $Q$ is a twice-contravariant symmetric tensor field. It is equivalent to giving a bundle morphism $\beta: T^*Q \to TQ$ such that $\beta$ corresponds to the adjoint $\beta^*$ under the natural isomorphism $T^{**}Q \cong TQ$. Fine. Then he says that a sub-Riemannian structure uniquely determines a cometric subject to the conditions
(i) ${\rm Im}(\beta) = \mathcal{H}$;
(ii) $p(v) = \langle \beta(p),v\rangle$ for all $p \in T_x^*Q$, $v\in \mathcal{H}_x$, $x \in Q$.
Here's what I understand: given $p \in T_x^*Q$, since we require $\beta(p)$ to be horizontal, it suffices to know what $\langle \beta(p),v\rangle$, is for all $v \in \mathcal{H}_x$, as $\langle\cdot,\cdot\rangle$ is non-degenerate, so requiring such product to equal $p(v)$ defines $\beta$, makes it as smooth as $\mathcal{H}$, takes care of (ii), and already gives us ${\rm Im}(\beta) \subseteq \mathcal{H}$.
I can't see why actual equality ${\rm Im}(\beta) = \mathcal{H}$ should hold. Namely, I don't see a completely natural way to produce $p \in T_x^*Q$ such that $\beta(p)=w$, where $w \in \mathcal{H}_x$ was arbitrary. Obviously condition (ii) defines $p$ on $\mathcal{H}_x$, but how to extend it? One can take an auxiliary Riemannian metric on $Q$ and let $p$ annihilate $\mathcal{H}_x^\perp$, but then uniqueness of the cometric is not clear and it feels like cheating.
What am I missing?
P.S.: I was sure this forum had a sub-Riemannian geometry tag but I'm on mobile and can't find it. Whatever.