Four Layers, Two Neurons Each

A network reads a single number $x$ between $0$ and $1$. It has four hidden layers of two ReLU neurons each, a left neuron and a right neuron in every layer (eight hidden neurons in all), and one output neuron with no activation. Every weight and bias is fixed:

For example, the right neuron of hidden layer 3 outputs $\text{ReLU}(2 \cdot \text{left} - 4 \cdot \text{right} - 0.5)$, where left and right are the two outputs of hidden layer 2.

The network labels $x$ positive when its output $z > 0$ and negative otherwise. As $x$ sweeps from $0$ to $1$, the positive labels come in separate stretches, each cut off from the next by values of $x$ labelled negative.

Now picture a wide network with the same budget. It has eight ReLU neurons in a single hidden layer, each reading $x$ directly, and they feed one output neuron of the same kind. Its weights and biases can be anything you like.

Which statement is correct?