slop-catalog

damu — the AI-slop catalog

This is the reference both damu modes read from. Prevent renders the Fix lines into a prohibition ruleset. Remediate runs each entry as a detection lens, using Detect for the fact-signal and Legit when as the hard exclusions a lens must not flag.

The one rule that governs all the others

Every tell below is sometimes correct. A fintech dashboard is mostly cards. A children's app should be rounded and emoji-rich. A brand site may earn a parallax hero. The slop isn't the pattern — it's the pattern applied by default, without a reason, to everything. A finding is only real when the choice looks unmotivated and uniform. When in doubt, rate it LOW confidence rather than HIGH — surface it, but don't treat a plausibly-intentional choice as a confident finding. Flagging a justified choice as if it were obvious slop is itself a kind of slop.

A second governing read: soul. Slop is what you get when every decision defaulted. The opposite isn't "more design" — it's evidence that a human made a choice here and not somewhere else: one deliberate accent, asymmetry that means something, type that has a job, whitespace that breathes unevenly. Absence of any such choice across a whole page is the strongest tell of all.

---

SLOP-01 — Font chaos

• The tell: 4–5 competing typefaces, often a display serif + a geometric sans + a mono + whatever the component library shipped, with no hierarchy logic. Looks amateur because nothing is in charge.
• Why it reads AI: models stitch type from many sources and never cull. A human picks one or two and commits.
• Fix: one or two families, max. A type system (one display + one text, or a single superfamily with weights). Hierarchy by size/weight/spacing, not by swapping fonts.
• Detect: count distinct rendered font-family stacks across text nodes. >3 non-icon families ⇒ flag.
• Legit when: an intentional editorial pairing (display + body + a deliberate mono for code/data) — 3 families with a clear job each is a system, not chaos.