Docs that read like a page,
not a stack of blocks.

This paragraph is live — select it. Every line here is laid out by pretext, a text-measurement engine that computes line breaks with pure arithmetic instead of asking the browser to reflow. Because the engine knows exactly how wide each line can be, it can narrow a line to slip past an obstacle and widen it again once the obstacle ends. The orb to the right is a layout obstacle, not a floated image hack — the text genuinely flows around its bounding box, line by line, the way a magazine sets type around a photograph. None of this touches getBoundingClientRect or paints text to a canvas. The words you are reading are ordinary, selectable, screen-reader-friendly DOM text; pretext only decided where each line should sit.

Documentation is not just prose — it is prose interleaved with examples, diagrams, and asides, and in a conventional renderer each of those interrupts the reading flow. With a measurement-driven layout, the code sample sits in the margin while the explanation keeps flowing beside it, so your eye never leaves the paragraph to find the example it describes. The block on the left is exactly that: a real, syntax-highlighted sample pinned to the column edge with the text wrapping cleanly around it. Same Markdown you already write — rendered as a page someone actually wants to read.