Methodology

What we monitor

We monitor a rotating set of high-traffic Reddit communities where retail stock discussion concentrates, plus public ticker streams on StockTwits. The exact board list and endpoints change as sources evolve. We keep the public site vague on purpose so the product stays about aggregated signal, not a map to raw feeds. New posts are pulled on a fixed cadence through the official Reddit API, with archival pulls where allowed for history. StockTwits pulls run on their own schedule. Additional sources may be added as we expand coverage.

How posts are analyzed

Each post runs through a structured pipeline. Tickers are extracted and normalized. We also score how upbeat or downbeat the wording is on a scale from -1 to +1 using a lightweight model (Gemini Flash 2.0) with deterministic post-processing. That score is a read on language, not a rating of the security and not trading advice. We surface recurring topic keywords so briefs stay anchored to what people typed, not a one-word market call.

Separately, each author gets an account-quality score. Inputs include account age, karma level, plus how broadly posts show up across subs. Low-quality or bot-like accounts reduce the weight of their mentions in aggregate views.

Coordinated pump detection

Coordinated promotion leaves footprints. We combine several heuristics. No single score decides on its own. They include:

• Account age versus posting velocity
• Comment-to-post ratio anomalies
• Cross-subreddit posting patterns in short windows
• Time-of-day clustering relative to baseline
• Karma trajectory versus typical organic curves
• Linguistic similarity to other recent posts pushing the same tickers

These heuristics flag suspected coordinated activity. Flagged posts are still stored for audit, but excluded from trending calculations so the surface signal stays cleaner.

How trending is calculated

A ticker is treated as trending when its mention count in the last 6 hours exceeds 200% of its 7-day rolling baseline, with a floor of at least 8 mentions from at least 4 unique non-flagged accounts. Baselines adapt as liquidity in the name changes so mega-caps are not flattened against microcaps.

What we report

Daily briefs summarize top movers and dominant topic keywords. Rationale lines are paraphrased from public posts. Trending updates list tickers crossing the surge threshold with mention deltas and time windows. The agent never adds buy or sell language. It does not attach price targets or personal guidance.

What we do not do

• We do not issue buy or sell recommendations
• We do not generate price targets
• We do not give personal advice
• We do not trade ahead of subscribers
• We do not sell user data
• We do not track which tickers individual subscribers privately watch in order to build positions

Limitations

The agent is useful for detecting mention trends early. It does not predict which trends become price moves. We do not claim it does. Sophisticated coordinated actors can still evade heuristics for a window. Output quality depends on what is publicly readable. Private chats, paywalled rooms, and offline flow are invisible.

How this page changes

When methodology changes, we update this page and post a short methodology note to Discord. Material changes are versioned in the repo that powers the agent so there is a paper trail.