# The twenty indicators One paragraph each: what it looks like, why models produce it, and when it is fine. Names match `knowledge.INDICATORS`. An indicator is a tell, not a rule: it counts as a default when it appears without a reason, and as a choice when it repeats consistently and connects to the subject's world. ## Type ### Instrument Serif An elegant, slightly condensed display serif, usually in italic, set large over a dark or gradient ground. Models reach for it because it was the tasteful choice on thousands of 2025 landing pages, so it became the statistically safe way to look designed. The same goes for Playfair Display, Fraunces and DM Serif Display when they arrive without a reason. It is fine when the subject is editorial, literary or luxury and the face is part of a type system the rest of the page follows. ### Excessive italics One or two words in a headline set in italics for flavor, as in Plans *first*. Acts *second*. Models imitate editorial emphasis without having anything to stress, so the italics land on arbitrary words. Italics are fine for one word of genuine stress, a title, or a term being defined. ### Negative kerning Tight negative letter-spacing on every heading and often on body text, until letters touch. Models learned that display type is tracked tight and applied it everywhere, including small sizes where it hurts reading. A slight tightening on large display type is normal; body and small text should keep the font's own spacing. ### Excessive mono fonts Monospace used for labels, captions, nav items, stats and taglines to feel technical. Models associate monospace with developer credibility and sprinkle it as costume. It is fine inside code, terminal output, data tables and anything a person would actually type or read as code. ### ALL CAPS subtitles A small uppercase line of adjectives under a headline, often letter-spaced and faded. Models produce it as a generic sign of polish; nobody reads it, and uppercase at small sizes is hard to scan. Short uppercase labels are fine where they carry real state or category, such as Sold out, or as a disciplined small-caps system in a brand that owns it. ### Periods in titles Short staccato headlines with full stops. Fast. Confident. Generic. Models copy the rhythm of famous campaign lines without the specific claim that made them work. A single sentence headline with a period can be fine when it is specific and the voice is consistent across the page. ## Surface ### Gradient backgrounds A diagonal purple to blue gradient, often with a touch of cyan, behind a hero or a card. It is the most recognizable AI background because it was the default of every template a model trained on. A gradient is fine when its colours come from the subject, such as a sunset for a travel brand, or when it is subtle grain over a committed ground colour. ### Color shades and opacities Many tints of one hue at 10, 20 and 40 percent opacity, stacked on each other until nothing has a clear edge. Models avoid committing to a palette and fake hierarchy with transparency instead. A couple of alpha steps for hover and disabled states are fine; a system of named tokens is better than a stack of rgba values. ### Squircles Large soft corner radii on every card, button and image, with nested corners that do not line up. Models default to roundness because it reads as friendly, and they rarely compute the inner radius. Rounded corners are fine when the radii form a small scale and nested ones follow inner radius equals outer radius minus padding. ### Containers with borders Every block boxed in a thin border, often inside another bordered box. Models wrap things in borders because a box is the simplest way to show a group; spacing would do it better. A border is fine on an input, a table or one card that must read as a distinct object. ### Overusing dividers Hairline rules between every section and inside every card. Models separate things with lines because it is explicit, while whitespace separates more quietly. A divider is fine where two things of the same kind sit close and must be told apart, as in a long list or a table. ### Dithering Pixel dither patterns added as retro texture regardless of the brand. Models copy it from developer-tool aesthetics as a shortcut to craft. It is fine when the subject is games, retro computing, print or zines, and the dither is used with intent rather than sprinkled on. ### ASCII images Pictures drawn from characters, used to look technical. Models produce them because they are easy to generate in text and read as developer culture. They are fine when the product is a terminal tool and the art means something, or when the subject is itself text. ## Structure ### Tags and pills Rounded badges on headings and cards that carry no real information: New, Beta, AI-powered on every item. Models add them because templates had them. A pill is fine when it shows real state that changes, such as Sold out or Draft, and appears only where that state applies. ### Eyebrows A tiny uppercase label directly above a heading, usually repeating it. Models produce it as a layout habit inherited from magazines, where it named a section. An eyebrow is fine when it tells the reader something the heading does not, such as the section of a long document or a date. ### Section numbering 01, 02, 03 on content that has no order. Models number sections to imply structure without having any. Numbers are fine when the order matters, as in steps or chapters, and the numerals are set as data rather than decoration. ### Code breadcrumbs File paths, arrows and comment syntax used as decoration on a marketing page: src/app/page.tsx with an arrow to nowhere. Models borrow the look of a terminal to feel credible. They are fine when the path is real and shown as part of the actual product or output. ## Content ### Verbose, redundant text Copy that restates the headline, stacks adjectives and says seamlessly, powerful, empowering. Models pad because length looks like effort, and a generic sentence is safer than a specific claim. Copy is fine when each sentence adds something a reader needs and could not have guessed from the heading. ### Over-the-top detailed UI Fake dashboards crammed with tiny charts, numbers and status dots to look impressive. Models fill space with detail because an empty area reads as unfinished. Dense interfaces are fine when they show the real product and its real data, and the detail is what the user came to see. ### Overly detailed diagrams Flowcharts with more boxes, arrows and glows than the idea needs. Models illustrate everything they can name, so the diagram grows past the concept. A diagram is fine when every node and arrow carries meaning and removing one would lose information.