Sentiment methodology

The Sentinel reports a social sentiment read. It's worth understanding exactly how that number is produced.

Lexicon-based, not a black box

Sentiment is scored in code, not by the language model. Pyxis matches posts against fixed sets of positive and negative words and emoji, then computes a breakdown. The agent interprets that breakdown in plain language, but the score itself is deterministic and reproducible.

What feeds it

Twitter activity (within a strict per-run budget), Reddit discussion, the Fear & Greed Index as a macro overlay, and — for protocols — Snapshot governance proposals.

Next: What Pyxis can & can't do →