Readability, and which sentence is the problem PLAIN

Readability scores have a bad reputation and mostly deserve it

They count syllables and clause lengths. They cannot tell whether a sentence means anything. A page of confident nonsense scores beautifully, and a careful technical paragraph scores badly because the words it needs are long and there is no shorter word that is also correct.

Anybody who writes to hit a number is making their writing worse. The predictable result is short sentences chopped where the meaning wanted to continue, technical terms replaced with vaguer ones, and connective words (because, although, whereas) removed, which is how an argument stops being an argument and becomes a list of assertions.

Four scores are shown rather than one, and they are shown together because they disagree. Where they disagree the disagreement is itself the information: it usually means the text has an unusual shape (long words in short sentences, or the reverse), and which of those it is changes what you would do about it.

What they are genuinely good at

One thing: finding the sentence that is too long. Not the average, the specific one. The forty-eight word monster in the middle of paragraph six, with three subordinate clauses and a parenthesis, that the writer cannot see because they know what it means and can therefore read it without effort.

That sentence is where readers stop. Not consciously; they reread the first half, lose the thread, skip to the next paragraph, and arrive at your conclusion having missed the reason for it. And it is invisible from the inside, which is exactly why a machine that cannot understand a word of it is useful here.

So underneath the scores is the half worth acting on: the longest sentences listed worst first, so you fix the one that matters and not the average; the passive constructions; the nominalisations, which are verbs turned into nouns, and which take the actor out of the sentence with them (“a decision was taken regarding the implementation” has no one in it at all); and the long words where a short one exists.

How to use the list without wrecking the prose

Treat every item as a question, not an instruction. Take the longest sentence and read it aloud; if you run out of breath or lose your place, split it at the join the meaning already has. Do not split it in half because it is long.

The passive voice is the one people get most wrong. It is not a fault. “The samples were collected in March” is right when who collected them does not matter. It is a fault when it is hiding somebody: “mistakes were made”, “it was decided”, “the account was closed”. Ask who did it, and if the answer is deliberately missing, that is a finding about the document rather than about its grammar.

Long words are the weakest signal of the four. Utilise should be use; anticoagulant should be anticoagulant. The test is whether a shorter word means the same thing, and only you know that.

Nothing you paste here leaves the tab. For the neighbouring problems, HOUSESTYLE checks a document against your own rules rather than a generic score, CADENCE measures the habits that identify you as the writer, and BRIEF is about structure, which is the thing no sentence-level measure can see.