Understanding Stop Slop's 5-Dimension Scoring System: A Technical Deep Dive

Stop Slop evaluates prose quality across five independent "S-dimensions" (Directness, Rhythm, Trust, Authenticity, Density), each rated 1-10 and summed to a maximum of 50, with any total below 35 flagged for mandatory revision.

The hardikpandya/stop-slop repository provides a structured framework for identifying and correcting low-quality AI-generated text through its quantitative 5-dimension scoring system. Defined in SKILL.md (lines 50-58) and documented in the README (lines 42-55), this methodology transforms subjective editorial instincts into measurable metrics. The system enables writers and automated tools to detect "slop"—characterized by hedging language, filler phrases, and metronomic rhythms—by assigning numeric scores that drive specific revision workflows.

The Five S-Dimensions Explained

Stop Slop’s evaluation framework centers on five qualitative attributes, each assessed through a specific diagnostic question. These dimensions are designed to isolate distinct markers of robotic or low-quality prose.

Directness

Directness measures whether a sentence states facts or instructions outright versus hedging with introductory clauses or softening phrases. The evaluator asks: "Statements or announcements?" According to SKILL.md, high-scoring prose eliminates throat-clearing constructions like "It is important to note that..." and delivers information immediately. Scores in this dimension range from 1 (heavy hedging) to 10 (immediate, unvarnished statements).

Rhythm

The Rhythm dimension evaluates sentence-length variation to detect metronomic patterns typical of AI generation. The diagnostic question is: "Varied or metronomic?" Text that alternates predictably between short and long sentences, or maintains uniform length throughout, receives lower scores. Human prose typically exhibits irregular cadences that break algorithmic patterns, scoring 8-10 on this scale.

Trust

Trust assesses whether the writer assumes the reader’s intelligence or resorts to over-explanation and soft-selling. The guiding question: "Respects reader intelligence?" This dimension penalizes excessive signposting ("Let me explain why this matters...") and unnecessary elaboration of obvious points. High-trust writing treats the audience as capable of inferring context without hand-holding.

Authenticity

The Authenticity dimension identifies AI-typical phrasing and mannerisms that signal non-human authorship. The evaluator asks: "Sounds human?" Markers include excessive adverbs, generic transitions ("Here's the thing..."), and overly balanced sentence structures. This dimension cross-references references/phrases.md and references/structures.md to identify specific constructions that degrade the score.

Density

Density quantifies removable filler content within the prose. The critical question: "Anything cuttable?" Unlike word-count metrics, this measures information compression—higher scores indicate lean text where every word serves a purpose. The references/phrases.md file provides a checklist of cuttable fragments (e.g., "very," "really," "in order to") that directly impact this dimension.

How the Scoring Algorithm Works

The scoring system operates on a summative aggregation model with a strict quality threshold.

Assessment and Aggregation

Each dimension receives an independent integer score from 1 to 10, creating a composite range of 5 to 50. As implemented in the skill definition, scores are simple summations without weighted coefficients:

Rate 1-10 on each dimension:
| Dimension | Question |
|-----------|----------|
| Directness | Statements or announcements? |
| Rhythm    | Varied or metronomic? |
| Trust     | Respects reader intelligence? |
| Authenticity | Sounds human? |
| Density   | Anything cuttable? |

(Extracted from [SKILL.md](https://github.com/hardikpandya/stop-slop/blob/main/SKILL.md#line-50‑58))

The 35-Point Threshold

The README establishes a non-negotiable quality gate: any total below 35/50 requires revision. This threshold ensures that no single dimension can mask deficiencies in others—a text scoring 9 on Rhythm but 4 on Authenticity (total 33) still fails and must be rewritten. The feedback loop directs writers to consult references/phrases.md and references/structures.md to address specific deficits before rescoring.

Practical Implementation Examples

The scoring system supports both manual editorial workflows and automated LLM integrations.

LLM Prompt Integration

To apply Stop Slop programmatically using Claude or similar models, structure the prompt to enforce the five-dimension rubric:

You are a style-checker using the Stop Slop skill.  
Rate the following paragraph on each S-dimension (1-10) and give a total out of 50.

Paragraph:
"The real challenge is that we haven’t yet built a robust pipeline. Here’s the thing: we need to double down on testing."

Scoring format:
Directness: X
Rhythm: X
Trust: X
Authenticity: X
Density: X
Total: X/50

Python Automation Helper

For batch processing or CI/CD pipelines, implement the scoring logic in Python:

# stop_slop_score.py

dimensions = ["Directness", "Rhythm", "Trust", "Authenticity", "Density"]
questions = {
    "Directness": "Statements or announcements?",
    "Rhythm": "Varied or metronomic?",
    "Trust": "Respects reader intelligence?",
    "Authenticity": "Sounds human?",
    "Density": "Anything cuttable?"
}

def ask_llm(paragraph):
    # Stub: replace with actual LLM call

    return {dim: 8 for dim in dimensions}  # dummy scores

def evaluate(paragraph):
    scores = ask_llm(paragraph)
    total = sum(scores.values())
    verdict = "Revise" if total < 35 else "Good"
    return {"scores": scores, "total": total, "verdict": verdict}

Running evaluate() returns a structured verdict indicating whether the prose passes the 35-point quality gate.

Source File Architecture

The scoring system is distributed across specific files within the repository:

File	Role
[SKILL.md](https://github.com/hardikpandya/stop-slop/blob/main/SKILL.md)	Core skill definition; contains the scoring table (lines 50-58) and quick-check rules.
[README.md](https://github.com/hardikpandya/stop-slop/blob/main/README.md)	Overview including the S-dimension scoring description and 35/50 threshold (lines 42-55).
[references/phrases.md](https://github.com/hardikpandya/stop-slop/blob/main/references/phrases.md)	Lists filler phrases and adverbs affecting Density and Authenticity scores.
[references/structures.md](https://github.com/hardikpandya/stop-slop/blob/main/references/structures.md)	Describes structural clichés impacting Directness, Rhythm, and Trust.
[references/examples.md](https://github.com/hardikpandya/stop-slop/blob/main/references/examples.md)	Before/after transformations for calibrating the scoring process.

Summary

• Stop Slop's 5-dimension scoring system evaluates prose across Directness, Rhythm, Trust, Authenticity, and Density.
• Each dimension scores 1-10, summing to a maximum of 50, with 35 as the minimum passing threshold.
• The scoring rubric is defined in SKILL.md (lines 50-58) and reinforced by the README (lines 42-55).
• Reference files (phrases.md, structures.md) provide actionable checklists for improving low scores.
• The system supports both manual editorial review and automated implementation via LLM prompts or Python scripts.

Frequently Asked Questions

What is the minimum passing score in Stop Slop's 5-dimension scoring system?

Any total below 35 out of 50 requires revision. This threshold is explicitly defined in the README (lines 42-55) and ensures that deficiencies across multiple dimensions cannot be masked by high performance in isolated categories.

How does the Density dimension differ from Authenticity?

Density measures removable filler and word-count efficiency ("Anything cuttable?"), while Authenticity detects AI-typical phrasing and mannerisms that sound robotic ("Sounds human?"). A text can be dense (few cuttable words) yet inauthentic (containing clichéd AI transitions), or authentic but bloated with unnecessary adverbs.

Which source file contains the official scoring rubric?

The canonical scoring table resides in SKILL.md at lines 50-58. This file defines the five dimensions, their diagnostic questions, and the 1-10 rating scale. The README references this rubric but does not duplicate the full implementation details.

Can the scoring system be automated with LLMs?

Yes. The system is designed for programmatic implementation using standard LLM APIs. By feeding the dimension definitions and questions from SKILL.md into a system prompt (as shown in the Python example), automated tools can return structured JSON scores and enforce the 35-point threshold in CI/CD pipelines or content management workflows.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}