Speed readers have modes. You pick one from a toolbar, it changes how the text arrives, and when you leave the page the mode goes with you. The mode is a property of your session.
Make it a property of the *document* instead give it a URL and it stops being a preference and starts being an argument.
The difference
`example.com/post/` and `example.com/post/?view=cadence` are two different readings of the same text, and I can send you the second one. "Here is my essay with the function words removed" is a thing I can now say with a link. So is "here is the shape of it, before you read a word": `?view=cadence` renders one bar per sentence and nothing else, and you can tell a staccato rant from a patient essay at a glance.
That only works if the server renders it. A mode that lives in a toolbar is invisible to a crawler, to a screen reader that has already flattened the page, and to a model fetching the URL. A mode that lives in the URL is just a document.
What a predicate is
A pure function from a token stream to a list of cells. It says one true thing about the words.
That definition is doing work. Because they are pure and total, they compose `?view=rare+bionic` runs one and then the other, and the result is still just cells. Because they are total, there is no view that fails on some document; the empty post renders as the empty view rather than a 500. And because they are pure, the same code runs at the edge, in the browser, and in `cargo test`, so the preview cannot disagree with the page.
The twelve
Four are ports of the reading modes at [read.mino.mobi](https://read.mino.mobi) bionic, RSVP, crawl, plain. Read is where the idea that a text has more than one arrival came from, and its dwell model (the pauses at sentence ends are the whole trick) is reproduced here rather than re-invented.
Read's *drills* used to be here too memorize, skeleton, spine. They are gone. This is a place to publish and to look at what you published, not to practise reading it, and a mode that quizzes you was always answering a question nobody had asked on this page.
The rest are analytic, and they are the ones that only make sense once a view is addressable:
**cadence** discards the words entirely and draws sentence lengths. Prose that is all one length reads as flat, and you cannot see that while reading it.
**grade** draws the same bars for reading *ease* rather than length, and the two disagree more than you would expect. A short sentence of Latinate abstractions scores worse than a long plain one; only one of the two views will tell you.
**hapax** weights each word by how rare it is *within this document*. Words used once burn; words used constantly fade. Your tics become visible.
**rare** weights by rarity in English at large instead the counterpart, and routinely disagreeing. A word used once here burns in hapax even if it is "house"; a word repeated twenty times still burns in rare if it is "sublimate". One finds what you do not repeat, the other finds where you reached outside ordinary English. Typos light up too, which is a feature I did not plan.
**sentiment** colours the words a valence lexicon rates. **emotion** lights up one emotion at a time `?emotion=fear` because eight colours of scattered words are indistinguishable to a colour-blind reader, and "where is the fear in this" is the better question anyway.
Both are descriptions of *vocabulary*, not readings of tone. The lexicons tag word forms out of context: negation is not modelled, so "not terrible" reads as terrible, and irony is entirely invisible to them. Worth having for what they actually are.
**concordance** alphabetises every content word with its neighbours the book as its own index.
**reverse** puts the last sentence first. Editors have read drafts backwards for a century, because momentum hides bad steps, and reading against the grain removes the momentum.
The honest limitation
Predicates operate on words. There is no meaningful cadence of a table, and `?view=rsvp` over a code listing is a punishment rather than a reading mode. So fenced code is skipped at tokenisation, and any view other than `plain` renders the token stream rather than the document structure: no headings, no lists.
That is a real loss and it is the right trade. A view is a lens, not a replacement. `plain` is always one click away, and it is the default.