SlopStopper 🛡️

SlopStopper is a local, privacy-focused automated system designed to audit a child’s YouTube history. It acts as an intelligent "guardian" that uses LLMs (Google Gemini) to detect "brainrot," radicalization pipelines, dark patterns, and low-quality content ("slop").

Unlike generic brand safety tools, SlopStopper adopts the persona of a cynical, protective parent who analyzes content intent rather than just keywords.

Features

• Local & Private: All data is stored locally in a SQLite database (data/slopstopper.db).
• 5-Dimension Analysis: Uses Gemini 2.5 Flash-Lite to evaluate content across Visual Grounding, Taxonomy, Narrative Quality, Cognitive Nutrition, and Risk.
• "Cynical" Persona: Detects "sigma male" rhetoric, "mascot horror," and dopamine-loop editing styles.
• Deep Dive Inspector: A "Content Fingerprint" view for individual video analysis, showing structural integrity, weirdness, and emotional volatility scales.
• Nutritional Scoring: A comprehensive 0-10 "Quality Score" for every video, evolving over time.
• Actionable Audit: Interactive "Kill List" and "Risk Spotlight" (Brainrot/Aggression/Slop) to isolate and block toxic channels.
• High-Fidelity Dashboard: A persistent, state-preserving Streamlit interface with "Diet" (Overview), "Audit" (Action), and "Deep Dive" (Inspection) tabs.

Prerequisites

• Python 3.12+
• uv (for dependency management)
• Google Gemini API Key

Setup

1. Clone & Install

   git clone <repo>
   cd slopstopper
   uv sync
   

2. Environment Variables Create a .env file in the root directory:

   GEMINI_API_KEY=your_actual_api_key_here
   

3. Data Place your YouTube watch-history.json (from Google Takeout) in the root directory of the project.

Runbook

1. Ingest Data

Parse your raw watch-history.json into the local database. This step is idempotent and can be run multiple times as you add new history files.

uv run src/ingest.py

2. Run Analysis

Analyze pending videos using the Gemini API.

Analyze specific videos (good for testing):

uv run src/analyze.py --ids VIDEO_ID_1 VIDEO_ID_2

Analyze a batch of videos (e.g., first 50):

uv run src/analyze.py --limit 50

Analyze ALL pending videos:

uv run src/analyze.py --all

Compare Models (A/B Test with Judge):

uv run src/compare_models.py VIDEO_ID

3. View Dashboard

Launch the local web interface to explore the results.

uv run streamlit run src/report.py

4. Run Tests

Execute the test suite (includes mocked LLM interactions).

uv run pytest

Directory Structure

• src/ingest.py: Parses JSON to SQLite.
• src/analyze.py: Main logic for transcript fetching and LLM analysis.
• src/report.py: Streamlit dashboard.
• src/schema.py: Pydantic models collecting the analysis schema.
• src/prompts.py: System prompts defining the "SlopStopper" persona.
• data/: Stores the SQLite database.