Get page numbers that contain rotated text

In that case looking at the text quads is probably the only option, but it should be sufficient to look at a single word in each page.

Assuming that all the text on a page is oriented the same way, there is no need to loop over all the text in the document. The "quads" are the four points that mark the corners of the bounding box around the word. Create a document with four pages, with every page using a different text orientation. Then run the following code in your JavaScript console:

var q = this.getPageNthWordQuads(0, 1);
console.println(q.toSource());
q = this.getPageNthWordQuads(1, 1);
console.println(q.toSource());
q = this.getPageNthWordQuads(2, 1);
console.println(q.toSource());
q = this.getPageNthWordQuads(3, 1);
console.println(q.toSource());

This will print the quads for the first word on all four pages. When you now analyze the four points in each quad, you will find that they are arranged in a certain way (this assumes that you use a word with a width that is greater than it's height). Let's assume you get this output (simplified to integers):

[[85, 744, 115, 744, 85, 728, 115, 728]]

This means that P1 has an x component of 85, and a y component of 744, P2 is 115 and 744, and so on. You can see that the y components of P1 and P2 are the same, and P1.x is less than P2.x. You do this for all four pages, and you will have rules to find the text orientation on each page. To make sure that you don't accidentally pick a first word like "I", I would not just use one word, but maybe the first 3 and use a 2 out of 3 rule to determine the text orientation.