How the sausage is scored

Methodology

The original ask from FinTwit: a CNN Fear & Greed / AAII-style reading that only watches retail chatter on X. This build is a complete product shell with a deterministic scoring engine so it deploys without API keys. Swap the seed tape for a live firehose later.

The five legs

1. Language (28%). Lexicon of bull and bear phrases scored per post. Matches sit on word boundaries and the longest phrase wins, so “cut rates” counts as easing and does not also count as the bear word “cut”. “top” does not fire inside “desktop”. The number on a post is that lexicon reading.
2. Engagement (22%). Likes, reposts, replies, and follower count via log weights so mega-accounts do not nuke the tape.
3. Breadth (18%). Share of posts that resolve bull vs bear. A melt-up with only three accounts yelling is not breadth.
4. Desk lean (18%). Named FinTwit desks with a stance and a stylized hit rate.
5. Velocity (14%). Time-of-day pulse so the number is not a statue.

Buckets

• 0–19 Extreme Fear
• 20–39 Fear
• 40–59 Neutral
• 60–79 Greed
• 80–100 Extreme Greed

What this is not

Not investment advice. Not a live X API product yet. Posts are seeded archetypes of the timeline — flow accounts, perma-bears, delivery bulls, headline bots, and the guy going 3x QQQ into the close.