Prose Layout Detect

The bug this exists to prevent

Any surface that renders user- or model-supplied text eventually collapses whitespace, because that is what HTML does and what a paragraph renderer should do. Then someone pastes a measurement table, a code sample, or aligned query output, and it arrives as a single wrapped line with the columns shuffled into the words.

Nothing errors. The text is all there. It is simply no longer readable, and the person who wrote it cannot tell why their table looks like a sentence.

The naive fix — preserve whitespace whenever there is more than one consecutive space — is worse, because ordinary typing contains double spaces after full stops, and now every other paragraph renders in a monospace block.

What counts as layout

Positive cases, each with its own test: an aligned measurement table, an indented code sample, a single-line table header, and SQL output with aligned columns.

Negative cases, which are the ones that make it usable: ordinary prose is left alone, a **single double-space is treated as typing** rather than structure, and a markdown-style bullet block is not preformatted — bullets are semantic markup that a renderer should handle properly, not whitespace art to be frozen.

Why it lives in its own module

Deliberately separate from the component that renders it. The component is .tsx, and importing JSX into a .test.ts puts the suite in the wrong test project — so the rule that actually needs testing would have been the one part that could not be tested.

A one-line predicate with seventeen tests beside it, rather than an if inside a component with none, is the whole point.

What this does NOT do

No rendering, no markdown parsing, no syntax highlighting, no sanitisation. It answers one question and returns a boolean.

It is heuristic. Text that is genuinely ambiguous — a poem, an ASCII diagram made of prose — will be judged one way or the other, and the tests document which.

Verified

17 tests across aligned tables, indented code, single-line headers, aligned SQL output, ordinary prose, the single-double-space case, and markdown bullets.

Not covered: no test on non-Latin scripts, where alignment by character count does not correspond to visual alignment at all.