What we do to every source
- Load the publication’s homepage in a real browser: not a fetch of the HTML, because most of what an ad-heavy page brings in arrives after the first response.
- Accept the cookie banner. This is the step that decides whether the exercise is honest. Consent changes measured ad load by up to twenty-fold and reverses the ranking: one major news site loads two ad entities with the banner up and forty-three after accepting. A source whose banner cannot be cleared is flagged, never scored.
- Count the distinct ad entities and third-party domains contacted, record layout shift (CLS) and how much of the first screen an interstitial covers, and take a full screenshot.
- Look up reach as a Tranco rank for the registrable domain. Where the source is a section or a subdomain rather than the site itself, the rank is treated as an upper bound inherited from the parent, and the page says so.
A bot wall answers HTTP 200 and loads almost nothing, so a challenge page would score a perfect ad number. Those are detected and excluded rather than counted.
Measured
Derived from what the page actually did when we loaded it. The raw numbers behind these are printed on each source’s own page, with the date they were taken.
- Ad restraint
- Ad entities counted on the homepage after accepting the cookie banner
- Reach
- Tranco rank for the registrable domain, capped where the feed is a section
Judged
Our opinion, applied consistently across the set and stated as opinion. These are not measurements and we do not present them as ones.
- Design
- How calm the page looks — clutter, hierarchy, typography, whitespace — judged by a vision model from our screenshot and corroborated by measured reading comfort. Recalibrated 2026-08-31 so the catalogue median sits at 6; a plain page that reads well scores high here, by design
- Editorial
- Curated judgement of the writing, not a measurement
- Signal-to-noise
- How much of the feed is the thing you subscribed for
- Access
- How much is readable without a paywall or a login
- Cadence
- Whether it publishes often enough to be worth a slot
- Independence
- Ownership, and how much it constrains what gets covered
How the score is built
Each dimension is scored 0–10 and weighted into a single score. Editorial carries the most weight, ad restraint and reach the next; the remaining five weigh equally. The weights are published here rather than described, so the arithmetic can be checked against any source’s own page.
If yours is wrong
Measurements are of one page load at one moment, and sites change. If a number on your publication’s page looks wrong, or the screenshot has aged out of date, tell us and we will re-measure it. The capture date is printed on every source page precisely so that claim can be checked rather than taken on trust.