#The correction model
A good anti-slop review should improve decision quality while preserving the project's own visual direction.
#Step 1: recover the brief
Before changing the interface, identify:
- product purpose;
- audience;
- primary user task;
- desired emotional quality;
- existing design tokens;
- brand colors;
- typography choices;
- reference products or visual references;
- platform conventions;
- accessibility constraints;
- density requirements;
- domain-specific expectations.
When those inputs already exist, the agent should follow them.
When they do not exist, the agent should expose the missing decision and avoid silently choosing the statistically common answer.
#Step 2: find defaults masquerading as decisions
Look for:
- untouched starter tokens;
- repeated stock component combinations;
- default colors;
- default font choices;
- template page structures;
- generic icons;
- standard motion wrappers;
- filler copy.
The agent should identify the underlying default before proposing a visual replacement.
#Step 3: identify what each element is trying to do
Map each candidate element to a function:
- information
- hierarchy
- grouping
- affordance
- state
- feedback
- navigation
- accessibility
- identity
- emotion
An element with no function becomes a removal candidate.
#Step 4: remove before decorating
Many weak interfaces improve when unnecessary material disappears.
Try removing:
- one container level;
- one description;
- one badge;
- one icon wrapper;
- one decorative accent;
- one animation;
- one redundant heading;
- one generic CTA;
- one fake metadata label.
Then reevaluate hierarchy.
#Step 5: replace with project-specific choices
When an element still needs treatment, use the project's actual system.
Examples:
- apply the documented brand primary when the current color is an unspecified fashionable default;
- use documented radius tokens;
- choose layout based on content hierarchy;
- use real screenshots or data where product proof matters;
- write copy from real capabilities;
- use semantic status colors;
- connect motion to state.
#Step 6: verify the full state model
A de-slopped default state can still hide generated behavior.
Test:
- responsive widths;
- long labels;
- empty data;
- dense data;
- loading;
- error;
- permissions;
- keyboard focus;
- reduced motion;
- disabled actions;
- partial results;
- offline or retry behavior where relevant.
#Step 7: run the specificity review
Ask whether the finished surface could be transplanted into an unrelated product with only the logo changed.
If the answer is yes, inspect where the product's real structure, language, content, or identity has been lost.
Specificity should come from real product differences. Arbitrary decoration weakens that specificity.