NoVo’s server has tools that describe what a crowd is saying. It also has tools that describe what traders hold. An agent sees them all in one catalogue, and a model writing a summary will blend them into a line about sentiment. The two kinds of reading need to stay apart.
The reader poll
get_crowd_sentiment returns a public bull and bear poll, with one vote per address per day. It is a poll of readers on one website. No money moved to produce it. It is an opinion tally, and NoVo’s own notes to agents say it must never be quoted as positioning or flow.
A fair sentence is that readers of the poll lean one way today. An unfair one is that the market is bullish.
The tag count
get_retail_chatter takes a symbol and reports on posts about it on StockTwits. Users there can tag their own post as bullish or bearish. The tool counts those tags. It reads one venue, and it measures what people chose to label.
The catch is the denominator. Only a fraction of posts carry a tag, and that fraction varies a great deal from one symbol to another. So the tool ships a self_tagged figure beside every count. Below eight tagged posts it returns no percentage at all. An agent should repeat the count of tagged posts with any share it quotes. A bullish share with no count behind it could be a handful of posts.
Short volume is not a vote
get_short_volume reports off-exchange short volume. It is tempting to read a high figure as bearish opinion. Most off-exchange short prints are market makers hedging. The honest reading is one-sidedness in the day’s off-exchange trading. An agent that calls it bearish sentiment has added a meaning the data does not have.
What positioning looks like
Positioning is what someone holds. get_futures_positioning returns the weekly Commitments of Traders figures, which show what speculators and commercial traders hold in futures. get_dealer_levels describes where options dealers are positioned on the index funds. These come from holdings and open contracts. Nobody had to post anything.
Even here the agent must be careful. Dealer gamma speaks to the size of moves and gives no direction, as set out in dealer levels are structure, not signals.
How the blend happens
Ask an agent how retail feels about a stock and it may call the chatter tool, the poll and the short volume tool. Then it writes one paragraph: retail is bullish, shorts are heavy, the crowd is split. Three readings of three different things have become one mood. None of the sources said that.
The instruction
Tell the agent to name the kind of each reading. A poll is a poll of readers. Chatter is tagged posts on one venue, with the count. Short volume is one-sidedness in off-exchange trading. Positioning comes only from the positioning tools. Do not combine them into a single view.
Then ask it about one stock and read the answer for the word sentiment. If that word appears with no source beside it, the blend has happened. The broader limits of reading mood from text are in sentiment analysis in trading.
The crowd tools are free on the MCP & API, and each carries its own warning in its description. They are there because a crowd reading has uses. It only has them when it is labeled as what it is.