How Wide Is the Street?

A bank trains a linear support vector machine to sort loan applications into approve and deny, using two features: $x_1$, the applicant's yearly income in units of $10,000, and $x_2$, the length of their credit history in years.

The trained model gives every applicant a score

$$s = w_1 x_1 + w_2 x_2 + b, \qquad w = (0.5,\ 1.2), \qquad b = -4$$

The decision boundary is the line $s = 0$. The two edges of the "street" around it are the lines $s = +1$ on the approve side and $s = -1$ on the deny side, and the support vectors sit exactly on those edges. The weight vector $w$ points straight across the street, at right angles to all three lines. For any vector $d$, the distance $d$ covers in the direction of $w$ is $\frac{d \cdot w}{\lVert w \rVert}$, where $\lVert w \rVert = \sqrt{w_1^2 + w_2^2}$ is the length of $w$.

Two of the training applicants are support vectors:

How wide is the street? That is, what is the distance from the $s = -1$ edge to the $s = +1$ edge, measured at right angles to them? Give your answer rounded to 2 decimal places.