The gauges at the top are filled up to each rating. Page Quality uses Google's slider: Lowest, Low, Medium, High, and Highest, with Lowest+, Low+, Medium+, and High+ as half steps; typical pages are Medium. When you give a query, Needs Met runs from Fails to Meet to Fully Meets. N/A means the guidelines assign no rating, for example when the page did not load or is behind a paywall.

Each graded judgment shows a badge, a row of shaded boxes, and a marker. The boxes are the levels from lowest to highest, shaded by how much probability Jev put on each, so the darkest box is the single most likely level. The marker is the average position across all the levels, weighted by those probabilities, and the badge is the level nearest that average, which is also what the rating rules use. Confidence says how concentrated the probability is. "Split reading" means the single most likely level differs from the badge; a person should look.

Yes or no judgments show the probability of yes. Flags such as harm or deception are problems when the bar is high; items such as "creator identified" are good when it is high.

The rating rules run in order: harm, deception, spam, and copied content force Lowest before anything else; a firm Low attribute gives Low; High and Highest require the guideline's High criteria; YMYL topics get more scrutiny and non-YMYL topics less. A warning counts only when Jev's reading is firm. One it could not confirm is listed under "Why not higher" for a person to check and holds the rating down: one keeps it at High or below, two at Medium+. Real strengths next to a mild problem keep a page in the Medium range. Reputation research still has to be done by a person and can move the rating either way.

Accuracy is checked statement by statement. Code picks the statements with figures, dates, and rules, and Jev judges each one against today's date. The ones it doubts are listed under Accuracy, and all of them appear under "What Jev read".