Slop Finder: a Chromium extension that flags AI-slop style on X, LinkedIn and Reddit

# Slop Finder

**Automatic local AI-slop style detection for X, LinkedIn, and Reddit.**

Slop Finder is a Chromium extension backed by a local [Laya-MLX](https://github.com/mizorewww/laya-mlx) helper. It watches social feeds as you scroll, scores post-writing style locally, and marks strong style matches with an animated tape overlay.

> The score is a **heuristic style-match score**, not proof that AI authored a post. Text-only AI authorship detection is inherently uncertain, so Slop Finder is designed to abstain when context is insufficient.

## What it does

- Scans new posts automatically while scrolling.
- Supports **X / Twitter**, **LinkedIn**, and **Reddit**.
- Runs inference locally on Apple Silicon through MLX.
- No cloud API, no external scraping service, no page re-fetching.
- Uses multiple style dimensions rather than a single "AI or human" classifier.
- Applies a conservative consensus score to reduce false positives.
- Adds an animated "AI SLOP" tape only when the score crosses the configured threshold.
- Side panel shows scan counts, queue state, latency, and recent matches.

## Detection logic

Laya-MLX assesses **low information content** and **formulaic rhetoric** using
explicit criteria. Both must support the result. Engagement bait is also shown
as a diagnostic signal.

Lexical formulas, repetition, repeated English/Russian conditionals and contrasts
provide a capped supporting contribution. Repeated long dashes only support
already agreeing signals. Lists, polished writing, or a single dash do not
independently convict a post. Quotes and code are excluded from these lexical rules.

Scores are **heuristic points out of 100**, not probabilities. The default marking
threshold is **65/100**. Short, title-only, truncated, conflicting, or failed
analyses