# jargAI — full site content

> jargAI predicts crowd behavior. Not fundamentals, not sentiment: the actions people are about to take, learned from the patterns in what they have already done. Its first research program, Project α, is in private research.

Website: https://jarg.ai
Tagline: Everyone's playing. Few see the board.
Status: Project α is in private research. Early partners can request access at https://jarg.ai/#access.

## What jargAI does

jargAI models what people will do next. It does not model what something is worth (fundamentals) or how people feel about it (sentiment). It models the actions people are about to take, learned from the patterns in what they have already done.

## Our approach: most models ask the wrong question

Two people can feel exactly the same and do opposite things. So jargAI skips the opinions and goes straight to the behavior.

- Fundamentals ask "What is it worth?" That information is slow, public, and already known by everyone who is looking. jargAI does not model this.
- Sentiment asks "How do people feel?" It is cheap to fake and easy to misread, and feelings often never turn into actions. jargAI does not model this.
- Behavior asks "What will people do next?" Habits, timing, and who copies whom. Actions leave patterns, and patterns repeat. This is what jargAI models.

## Try to be unpredictable (interactive demo)

Visitors tap left or right as randomly as they can. Before every tap, a small model locks in a guess based on the visitor's last few moves (it looks for repeated sequences of up to five moves). A coin flip would be right 50% of the time. The page shows a live share of moves predicted, compared with the 50% chance line. The demo runs entirely in the browser; taps are never sent anywhere.

## The math: a crowd is a system, and systems can be modeled

People are more predictable than they think. They show up at the same times, react to the same triggers, and copy the same people. jargAI turns those habits into numbers.

- Who moves first (cascades): one person acts, a few copy, then the crowd piles in. jargAI maps that chain and measures how fast it spreads. Measures: spread rate, depth, time to tip.
- Who copies whom (lead-lag): when one account's actions keep showing up in everyone else's a little later, that account is a leader. jargAI measures the delay. Measure: cross-correlation at lag Δt.
- When they act (routines): behavior runs on routines, the same hours, the same days, the same triggers. Routines are forecastable. Measure: hour-by-weekday seasonality.

The output is a probability, never an opinion: P(act | history), a number between 0 and 1. How likely is this group to act, and how soon? Then it is checked against what actually happened.

## The game: three kinds of players, three different moves

A crowd works like a Keynesian beauty contest. Winning is not about being right; it is about knowing what everyone else is about to do. So the question is never just "what are they doing?" but "who is doing it?"

- The early: acts before it is obvious, and keeps doing it on data they have never seen. Move: follow.
- The echo: shows up after the crowd. Loud, sincere, and late. Move: ignore.
- The ring: accounts that keep acting together, on the same thing, at the same time. Their enthusiasm is the tell. Move: fade (expect the opposite of what they push).

The catch: every edge fades once it is crowded. When the early get copied, they become the echo, so the model keeps re-learning who is still ahead.

## How it works: from a stray post to an automatic decision

1. Collect: public, user-generated content (posts, comments, replies, reviews) from all over the web, gathered around the clock through jargAI's own proprietary APIs and collection methods.
2. Filter: not all automation is bad. Some bots relay real news and alerts; others fake enthusiasm at scale. jargAI's algorithms flag both, so good bots are kept and labeled and bad bots are thrown out.
3. Store: each item is tagged with who posted it, when, how fast it spread, and whether a human is behind it, then loaded into jargAI's data warehouse.
4. Enrich: telemetry from other sources and offline data join the stream, so the models see far more than what people post.
5. Model: regression models learn which patterns show up right before people act. They train only on the past and are tested only on the future.
6. Score: every candidate gets a probability. AUC (area under the ROC curve) measures how well that ranking separates what happened from what did not. Only the top of the ranking gets flagged.
7. Act: each flag comes with a pain threshold, set before it starts. If the signal weakens past it, an automatic trigger cuts it loose. No hoping, no hesitating.

## Rigor: we try to break it before you do

Every prediction is scored by how well it ranks what really happened, on data it never trained on. A score of 0.50 is a coin flip. A number only matters if it survives these six rules:

1. Test on the future: models learn from the past and are tested on a later stretch they have never seen, over and over (walk-forward validation).
2. A locked vault: everything after a fixed date stays out of training and is only used to check whether a result holds up (holdout period).
3. No peeking: a signal only counts if it existed before the moment of the prediction. Anything recorded later is thrown out.
4. Beat the obvious: every data source has to beat a simple baseline. If it does not add anything on new data, it does not ship.
5. Honest error bars: confidence ranges resample whole days, because everything on the same day tends to move together.
6. Publish the misses: most patterns that look real turn out to be noise or something obvious. When that happens, jargAI says so.

## Glossary

- Behavior prediction: forecasting what people will do next from the patterns in what they have already done.
- Information cascade: a chain reaction in which one person acts, a few copy, and then many follow.
- Lead-lag relationship: one account's actions reliably show up in others' actions after a delay.
- Keynesian beauty contest: a game in which winning depends on anticipating what everyone else will choose.
- Walk-forward validation: testing a model only on data from after the period it learned from, repeated across many periods.
- Holdout period: a locked stretch of time never used for training, only for checking results.

## Frequently asked questions

Q: What is jargAI?
A: jargAI is a research company that predicts crowd behavior. Instead of modeling what something is worth or how people feel about it, it models what people are likely to do next, based on the patterns in what they have already done.

Q: What is Project α?
A: Project α is jargAI's first research program. It is in private research and not yet available to the public. Early partners can request access at https://jarg.ai/#access.

Q: How is behavior prediction different from sentiment analysis?
A: Sentiment analysis measures how people feel or what they say. Behavior prediction measures what they actually do: when they act, what triggers them, and who they copy.

Q: How is it different from fundamental analysis?
A: Fundamental analysis asks what something is worth. jargAI asks what the crowd will do next, and who will move first.

Q: How do you know the predictions are real and not luck?
A: Models are only tested on data from after the period they learned from, with a locked holdout period never used for training. Every data source has to beat a simple baseline, confidence ranges resample whole days, and results that turn out to be noise are reported as misses.

Q: Does the demo collect my data?
A: No. The demo runs entirely in the browser.

Q: How can I get access?
A: Leave your email in the request access form at https://jarg.ai/#access.