How hard is this to read?
A number for the prose, not a verdict on the piece. Here is exactly what it measures and what it doesn't.
What the reading score is
The reading score is a readability index from 0 to 100, shown with a plain-language label — High School, College, Academic, Graduate, Expert. Higher means harder to read: longer sentences, denser vocabulary, more clauses to hold in your head at once. A low score reads like a newspaper brief; a high score reads like a journal article.
The label and the number say the same thing two ways. They're a heads-up about the effort a piece will ask of you — nothing more.
How it's derived
Arc Codex averages four standard readability metrics — Flesch-Kincaid, Coleman-Liau, SMOG, and Dale-Chall — and projects that average onto the 0–100 scale. Four measures instead of one smooths out the quirks any single formula has with a given kind of writing.
One honest caveat about our sibling site, Huntaegis: it projects a single Flesch-Kincaid grade onto the same 0–100 scale rather than averaging four. The labels mean the same thing on both sites — the inputs behind them differ. We'd rather tell you than let you assume the two numbers are computed identically.
What it's for — and what it isn't
The reading score describes the writing, not the reporting. It says nothing about whether an article is accurate, fair, important, or true. A dense, high-scoring piece is not better journalism than a plain one — it's just harder to read. A simple, low-scoring piece can be the more careful and more honest of the two.
A high score means dense prose. It never means better journalism.
The objectivity score
Alongside the reading score sits a separate 0–100 objectivity score. It measures how far the language leans toward opinion versus neutral description — a piece thick with loaded, persuasive wording scores low; one that mostly reports plainly scores high. Like the reading score, it annotates the prose and decides nothing about what you get to read. For corpus-wide patterns across everything Arc Codex has measured, see the live dashboard.
The measure only understands English. For a story published in another language we score the English translation shown on the page and say so under the dial. Translation can flatten tone, so those scores are softer than scores on text written in English.
View the live dashboardSponsored-content signals
Paid placements and vendor press releases often run as ordinary news. Under each story we count patterns they tend to share, in the source text, with the sentences that produced each count. The counts are worked out by fixed rules and word lists, not by a language model, so the same text always gives the same result.
Who is quoted. We find attributions (“X said”, “according to Y”) and class the speaker from the words around their name: company leadership or the vendor (chief executive, founder, “the company”) or an outside source (a researcher, an official, a university, a statistics agency). A story where most classed quotes come from the company is marked above the usual range.
Owning-to-renting language. Phrases that move a reader from buying to renting: subscription, as-a-service, pay as you go, no upfront cost. We report how many per 1,000 words.
Statistic, then product. A figure credited to an outside source (Gartner, Eurostat, the Bureau of Labor Statistics and the like), followed within two sentences by a product or vendor offered as the answer. That move borrows the source's authority for the pitch.
Where the links go. For stories collected from 28 October 2026 on, we keep the links inside the story itself (not the site's menus or share buttons) and count where they lead: public and research sources (government, universities, statistics offices, journals), one private company's pages, links carrying tracking or affiliate codes, and sales or sign-up pages such as a demo request or a pricing page. Many links to one company, or several tracking and sign-up links, is marked above the usual range.
Writing style against the publication. A placed piece is often written by someone other than the newsroom. For English-language stories we count how often about 150 common words (the, of, which, however…) and punctuation marks appear, and compare that with the same publication's earlier stories using Burrows' Delta, a standard measure from authorship studies. We say where the story falls among the publication's own stories, and mark it when it differs more than 95% of them do. We only compare once we hold at least 20 earlier stories from that publication. Columns, interviews and wire copy differ in style too, so this is the weakest of the signals on its own.
Before launch (October 2026) we ran the rules over about two hundred English stories already on Arc: one was marked on any signal. None of this says a story is sponsored. Trade reporting quotes executives and covers subscriptions; a press release can quote nobody. Read the sentences shown and decide for yourself.
A signal marks a pattern worth a second look. It never means the story was paid for.
How stories are chosen
These scores describe a story once it is published. The rules that decide which stories are published in the first place (the sources we read, the filters, the topic ranking) are listed, with today's settings, on a separate page.
How we choose what to publish