About Scandal AI

What We Do

Scandal AI monitors the AI industry for emerging controversies, heated debates, and significant disputes in real-time. We track social media, news sources, and community platforms to surface stories as they develop.

Data Sources

Content is crawled from four primary source types:

Twitter/X — timelines of key AI industry figures and trending search terms
Reddit — AI-focused subreddit RSS feeds
Hacker News — front page and new stories
AI news RSS feeds — 10+ curated publications

Analysis Pipeline

Every piece of content goes through a multi-stage pipeline:

Ingestion — content is deduplicated (SHA-256) and matched against known AI figures
Spike Detection — a 30-minute keyword co-occurrence buffer clusters related content into topics
Classification — Claude Haiku quickly determines if content represents a genuine controversy
Full Analysis — Claude Sonnet produces structured analysis: summaries, key points, party positions, timeline, and forecast

Noise Score

Each controversy receives a noise score from 0 to 100, computed from seven weighted factors:

Reach (20%) — total audience exposed
Engagement (20%) — interaction velocity (posts per hour)
Star Power (15%) — involvement of high-profile industry figures
Cross-Platform (15%) — spread across multiple sources
Duration (10%) — how long the story has been active
Polarity (10%) — how divided opinions are
Industry Impact (10%) — potential lasting effect on the AI field

A 7-day half-life decay ensures scores naturally decrease as controversies lose momentum.

State Machine

Topics progress through five states based on signal thresholds:

Emerging — initial detection of a potential controversy
Growing — content velocity exceeds 15 items/hour
Debated — spread across 2+ platforms
Major Controversy — noise score reaches 75+
Resolved — activity drops below threshold after cooldown period

Each transition has a cooldown (12–48 hours) before decay can step the topic back to a lower state.

AI Disclosure

Summaries, key points, forecasts, and party analysis are generated by Claude (Anthropic). All AI-generated content is based on sourced material from the data sources listed above. Forecasts are probabilistic assessments based on current signals — events may develop differently.