Market commentary, turned into structured intelligence.

Buzzberg tracks public market commentary and turns it into structured market intelligence: tickers, directional calls, source links, sentiment, and speaker track records.

Buzzberg aggregates public sources such as YouTube, Twitter/X, Reddit, Substack, SEC filings, market data providers, and macro data feeds.

AI models extract candidate trade ideas, identify tickers, classify direction and sentiment, summarize the thesis, and attach confidence signals where verification is available.

Confidence is an internal quality signal for how clearly the source supports the extracted idea. It is not a prediction of investment returns.

No. Buzzberg is an informational and educational tool. It does not provide personalized investment advice, suitability determinations, or trade recommendations.

AI extraction can be wrong, incomplete, delayed, or based on ambiguous source material. Always verify important claims against the original source.

Feed shows source items you follow. Ideas shows structured, AI-extracted directional calls across tracked sources.

A trade idea is one extracted ticker call from a source item, including the ticker, speaker, direction, confidence, source link, entry price, and supporting rationale. A thesis is the rationale behind the call. On speaker pages, thesis view deduplicates repeated mentions and shows the first opened ticker-direction thesis for that speaker; long and short calls on the same ticker are treated as separate theses.

Long means the source is bullish on the ticker. Short means bearish or explicitly positioned against it. Avoid means the source flags risk or says not to own it, without necessarily making a short call. Watch or Neutral means the mention is mixed, wait-and-see, or not clearly directional. Saved is different: those are your own starred ideas, not an AI stance.

Use the filter bar above the Feed or Ideas list. Sources, Channels, Tickers, and Feeds change the universe; Direction filters Long, Short, Avoid, or Others; Score filters by confidence; Saved shows your starred ideas; Long Short focuses the feed on actionable calls; and First Call shows first-time ticker mentions. The reset icon clears active filters.

The Leaderboard compares authors using calls opened within the selected Call window, evaluated at the selected Return horizon. Alpha Rank measures performance relative to the S&P 500 with an adjustment for sample size. Changing either selector can change the ranking; typing a name or using additional table filters does not recalculate the scores.

For each eligible call, Buzzberg compares its return with the return of SPY, an ETF tracking the S&P 500, over matching start and end dates. The differences are averaged across the author’s calls. For example, a call returning +12% while SPY returns +5% outperforms by 7 percentage points. A positive call can still lag the market: +3% versus +5% is 2 percentage points behind. These are illustrative examples.

The average difference is then adjusted for the number of evaluated calls. Smaller samples are pulled more strongly toward zero, so a short track record has less influence. With the same average outperformance of 7 percentage points, 10 calls give an adjusted result of 1.75 points; 90 calls give 5.25 points. The same rule moderates negative results. Authors are ranked by this adjusted result.

At least 10 calls with both call and benchmark data are required for Alpha. A smaller sample may still show Calls, Average Return and Win Rate, while Alpha remains blank. The calculation uses the selected Call window and Return horizon. Search and additional table filters keep the scores from that full comparison group.

Call Window selects calls by their opening date. A 90-day window includes calls opened during the last 90 days, regardless of whether the author discussed the same company before. YTD starts on January 1 of the current year; All time uses the full tracked history. It does not set how long a call is held.

The two selectors work together. With a 90-day window and a 30-day horizon, a call opened 60 days ago can qualify; one opened 10 days ago cannot yet. A 30-day window with a 90-day horizon normally has no eligible calls: none are old enough. Opening dates define the pool; the horizon determines which calls can be evaluated.

Return Horizon selects the measurement point: 7, 30, 90, 180 or 360 days after each call opens. With a 30-day horizon, Buzzberg measures the price change from opening to the end of that horizon. If the author closed the position earlier, the return is fixed at that earlier close.

For example, a LONG opened at 100 and valued at 110 at the 30-day endpoint returns +10%. If it was closed on day 12 at 106, its result is +6%, even if the price later reaches 110. For a SHORT, a fall from 100 to 90 gives a +10% modeled return. These are illustrative prices.

Even an early-closed call must reach the full selected age before joining the statistics: a call closed on day 12 enters a 30-day comparison only once 30 days have elapsed since opening. Calls missing the necessary prices are excluded. A longer horizon can therefore show fewer eligible calls.

Calls counts the eligible calls for which both the call’s return and the matching SPY return are available. Average Return is the arithmetic mean of those call returns. Win Rate is the share with a return above zero; a zero return is not a win. These columns use the same set of calls.

For example, returns of +10%, −4% and 0% give 3 calls, a +2% average return and a 33% win rate. This is an illustrative three-call sample; Alpha would remain blank because it needs at least 10 comparable calls. Calls that are too recent, lack prices or cannot be evaluated are not counted. Alpha’s detailed tooltip shows comparison coverage and missing-data reasons.

The Ideas and Posts columns and the overview counters cover all time. They can therefore exceed the number of calls evaluated for the selected window and horizon.

SPY is the passive alternative for both LONG and SHORT calls. Its return uses the New York closing price on the publication date, or the most recent preceding close within five calendar days when that date has no stored close. The comparison ends on the call’s actual price-evaluation date, including an earlier close.

Call returns remain in the instrument’s quoted currency. The comparison does not convert currencies or adjust for dividends or volatility. It describes historical modeled calls rather than an invested brokerage portfolio.

Performance and Holdings are two views. Performance replays a model portfolio through history and compares it with the S&P 500 over the selected period. Holdings shows its underlying tickers, prices and returns by author or fund.

Calls and 13F choose the source in either view. Calls uses Buzzberg’s AI interpretation of public statements. 13F uses positions disclosed by tracked funds in quarterly SEC filings. Neither Performance curve is a verified account statement from the author or fund.

Holdings can appear as cards or a list. Expand a list row to see up to ten tickers. Cards can be reordered; the list preserves that order. Switching the layout does not change the return calculation.

Mentions selects the tickers an author discusses most often; Recent selects their newest calls. Performance builds a model of up to ten equally weighted positions, using the latest recorded call to determine long or short. Mkt cap narrows the selection by company size. In Performance it filters an author’s existing ten positions; it does not necessarily replace excluded names with new ones.

The Performance period sets the date range of the chart. In Holdings, 1D, 7D and 30D show price changes over the selected window. First Call uses the recorded first-call entry; aggregate portfolios label this Avg Entry and use the contributing authors’ average entry. The return beside a Calls author is the average of the displayed tickers with available returns.

The top filter row applies to all author cards or rows. Controls inside an expanded author change that author’s view.

Report quarter selects a disclosed quarter-end portfolio. Latest uses each fund’s latest available report. A filing is published after the positions’ reporting date, so it is a delayed view of the fund’s holdings.

Performance models copying at the first eligible market close after the filing becomes public, rather than buying at the earlier quarter-end date. Entry is this model price, not the fund’s purchase price. The holdings drift with prices until the next eligible filing rebalance.

In Holdings, 1D, 7D and 30D set the return window; ALL starts from the selected report’s copy date. The fund’s headline return comes from its full modeled copy portfolio. Each ticker’s Return is its own price change over the chosen window. The list shows the ten largest positions; cards also let you view all positions or changes.

Performance follows a portfolio through time, including changes to its positions. Holdings describes the currently selected set of tickers or report. Averaging today’s ticker returns does not recreate the history of a portfolio that changed along the way.

For Calls, the return beside an author averages the available returns of the displayed tickers. For 13F, the fund return follows the full weighted copy portfolio, including model cash; it is not an average of the ten visible tickers. Changing a ticker sort order or switching Cards / List does not change that 13F portfolio return.

Holdings quotes and historical curves also refresh on different schedules. Check the displayed period and data date before comparing values.

Calls reflects AI interpretations of public content. The author may hold none of the named securities, and the AI can misread a statement. Coverage, late-processed material and ticker corrections can affect historical results.

13F covers only disclosed securities, not a fund’s complete portfolio or real cash balance. The copy model excludes option rows and buys only non-option long holdings with verified tickers and usable prices. Untradeable disclosed weight remains model cash at zero return; the remaining holdings are not scaled back up to 100%. Consensus Top 10 uses today’s tracked manager roster, which can bias its historical results.

The displayed backtests do not deduct trading costs. Missing prices appear as a dash. These are research models, not actual account returns or investment recommendations.

Signed-in users can link Telegram and receive alerts for selected feeds, depending on their notification settings. Email is used for transactional login codes.

Buzzberg includes an MCP server for assistant workflows. Availability and access may change while the product is in beta.