Slop Shield

AI-powered slop detector for Chrome. Measures information density and verifiable claims on any webpage using a local AI model — 100% private, nothing leaves your machine.

What is Slop?

Slop = low information density. It's text generated to occupy space rather than convey meaning.

Density = Verifiable Claims / Text Length

Slop Shield scores every page you visit from 0-100 and breaks down exactly why something is sloppy — predictability, vagueness, repetition, filler content, and false confidence.

How It Works

Slop Shield extracts text from the current page and sends it to a local 7B parameter AI model (Qwen via Ollama) running on your machine. The model acts as a "Constitutional Slop Judge" with encoded values:

• Falsifiability — Can the claims be checked against reality?
• Epistemic Modesty — Does it acknowledge uncertainty or hallucinate confidence?
• Information Density — Ratio of novel, verifiable claims to total words
• Lexical Precision — Specific language vs clichés and buzzwords
• Structural Honesty — Organized to inform, or to create an illusion of depth?
• Source Attribution — Does it reference verifiable sources?

Works on Wikipedia, YouTube, arXiv, and any webpage.

Installation

Step 1: Install Ollama

Download from ollama.com or install via Homebrew:

brew install ollama

Step 2: Pull the Qwen Model

ollama pull qwen2.5:7b

This downloads the ~4GB quantized model. You only need to do this once.

Step 3: Start Ollama

ollama serve

Leave this running in a terminal. Ollama needs to be running for AI-powered analysis.

Step 4: Install the Chrome Extension

1. Download or clone this repo:

   git clone https://github.com/llSourcell/SlopShield.git

2. Open Chrome and navigate to chrome://extensions
3. Enable Developer mode (toggle in the top-right corner)
4. Click Load unpacked
5. Select the SlopShield folder you just cloned
6. The shield icon appears in your Chrome toolbar

Usage

1. Navigate to any webpage
2. Click the Slop Shield icon in your toolbar
3. Watch the animated analysis — the score ring fills up as the AI evaluates the page
4. Click OVERLAY to see slop-heavy paragraphs highlighted directly on the page
5. Click RE-ANALYZE to run the analysis again

Score Ranges

Requirements

• Chrome (or any Chromium-based browser)
• Ollama running locally
• ~4GB free RAM for the Qwen 7B model
• No API keys, no accounts, no cloud — everything runs on your machine

FAQ

Q: What if Ollama isn't running?
A: Slop Shield automatically falls back to heuristic analysis (lexical diversity, filler word detection, n-gram repetition). It still works, just less accurate than the AI model.

Q: Can I use a different model?
A: Yes — edit background.js line 3 and change the MODEL constant to any model Ollama supports (e.g., llama3.1:8b, mistral:7b, gemma2:9b).

Q: Is my data sent anywhere?
A: No. All analysis happens locally on your machine via Ollama. Nothing is sent to any external server.

Q: Why Qwen?
A: Qwen 2.5 7B offers strong reasoning at a small size. It runs well on ~4GB VRAM and follows structured JSON output instructions reliably.

License

MIT