skills/slop-audit/references/pattern-taxonomy.md

# Curated slop taxonomy

This taxonomy combines the strongest recurring signals from the two upstream projects with additional contextual safeguards.

## High-confidence patterns

### Throat-clearing
Announcements before the point: generic "here is the thing" style openings, self-conscious honesty setups, or a sentence that says it will explain before explaining.

Default action: cut the announcement and keep the substance.

### Binary reframe
Mechanical negation followed by the supposed real answer, especially repeated `not X, Y` structures.

Default action: state the useful claim directly.

### Faux-insight setup
Language that frames the writer as seeing what everyone else missed before giving an ordinary point.

Default action: remove the prestige setup and test whether the claim stands alone.

### Negative reveal
Several statements about what something is not before revealing what it is.

Default action: state the positive claim first.

### Interpretive metadiscourse
Text that tells the reader what to notice, how important the previous line was, or how to interpret a point the prose already showed.

Default action: delete or replace with evidence.

### Unsupported authority
Vague attribution such as unnamed experts, studies, reports, consensus, or audiences.

Default action: name the source or remove the attribution.

### Importance puffery
Abstract claims that label a moment, launch, change, or result as major without showing a concrete consequence.

Default action: replace the label with the consequence.

### Fake-profound ending
A closing line that converts a concrete argument into a slogan, aphorism, cute reversal, or performative mic-drop.

Default action: end on the last concrete point or next action.

### Synonym cycling
Changing a clear noun to near-synonyms across adjacent sentences only to avoid repetition.

Default action: repeat the accurate term.

### Generic CTA scaffold
A predictable ending that asks readers to comment, agree, share, or "think about it" without a real next action tied to the piece.

Default action: remove it or use a specific next step.

## Contextual patterns

These are signals, not bans.

### Adverb clustering
Several weak intensifiers or hedges appear close together.

Keep an adverb when it changes meaning, carries voice, or creates deliberate emphasis.

### Passive voice
Flag when it hides the actor or responsibility. Keep it when the actor is unknown, irrelevant, obvious, or when technical conventions favor the passive.

### False agency
An abstract or inanimate noun performs a human decision or action in a way that hides who acted. Do not flag ordinary metaphors or conventional expressions automatically.

### Wh-word openings
Flag only when several sentences use the same explanatory template. One natural opening is not a problem.

### Rhetorical questions
Flag when the writer asks and immediately answers predictable questions as a structure crutch. Keep real questions that create tension, invite response, or fit the voice.

### Fragments and staccato
Flag clusters of short dramatic fragments used as manufactured emphasis. Keep intentional spoken cadence and copywriting fragments when they sound like the writer.

### Short-line stacking
Flag prose that is broken into a run of very short standalone lines or micro-paragraphs, especially when several consecutive lines carry only a few words and each is treated like a punchline. This often manufactures pace instead of earning it through syntax or meaning.

Do not flag short lines automatically in poetry, lyrics, scripts, dialogue, UI copy, headlines, captions, lists, tables, code, or a user-requested line-by-line format. In ordinary prose and social copy, treat four or more consecutive short standalone lines as a strong contextual signal, then confirm by reading the surrounding passage.

Default action: combine related micro-lines into natural sentences or fuller paragraphs while preserving real emphasis, spoken pauses, and intentional cadence.

### Em dash clustering
Flag repeated decorative use or a dash doing the job of every other punctuation mark. One well-placed dash is not automatically slop.

### Three-part symmetry
Flag repeated triads, repeated three-bullet patterns, or paragraphs with identical three-part architecture. A single three-item list is normal writing.

### Uniform paragraph geometry
Paragraphs repeatedly have the same length, sentence count, and punch-line ending.

### Abstract noun chains
A sentence stacks abstract concepts where a concrete actor, action, object, or consequence would be clearer.

### Specificity theater
The prose sounds precise but uses invented-looking categories, vague metrics, generic numbers, or unsupported examples. Never "fix" this by inventing specifics.

### Synthetic intimacy
The text imitates personal closeness, empathy, confession, or conversational familiarity that is not supported by the writer's actual voice.

### Style laundering
The draft removes corporate or model-like prose but replaces it with forced slang, profanity, quirky fragments, or self-conscious casualness.

## Voice-dependent signals

Preserve by default when they are characteristic and clear:

- colloquialisms
- profanity
- self-interruptions
- deliberate repetition
- fragments
- long spoken sentences
- hedges that express real uncertainty
- rhetorical questions with a genuine job
- personal asides
- unusual punctuation used sparingly

The audit should distinguish `generic pattern` from `writer habit` instead of treating both as the same defect.