Instead of this:

consider the following construction:

This way, the diagram is smaller, clearer and the data flow is more appearent.

Further, this selection VI is optimized and does not use a simple build array to construct the result. For example, selection 0.5 out of 1 million strings goes in less than 1 second on my Athlon 800MHz (only first run takes longer, LV does the allocations only once).