Anatomy · how a research note is built
Anatomy of an AI stock research note.
A research note is not a stock tip. It is a structured argument you can check line by line. Here is a live one, taken apart: every part, what it says, and which step and Claude model wrote it.
The example
This page takes apart APPN — Appian Corporation, one of the live research notes. The example changes over time; the structure below does not: every one of the 1,125 notes is built from these same parts.
-
Who the company is, and whether the portfolio owns it.
APPN → APPN · Appian Corporation · In the portfolio
How this part is made
Code keeps only US stocks that trade enough, by daily dollar volume and bid-ask spread. The status says whether the portfolio holds the stock or the research is watching it. It comes from what the portfolio holds, not from the research note.
-
One paragraph: the call, and why, in plain English.
APPN → Appian's third-quarter 2026 revenue rose 25% to $234.6 million and full-year guidance went up again, giving the shares a modest edge over the S&P 500; the fourth-quarter report, expected around 18 February 2027, shows whether cloud…
How this part is made
Claude reads the last two weeks of news and the price history and searches the web for the company's releases, filings and analyst actions. Then it sums them up as one claim that a future price or report can prove wrong. Not a price target: a reason.
-
The note's own test of what would prove it wrong: a price level, a business fact, or both, published before the market decides.
APPN → A daily close below $49.10, the 30 October 2026 close before the third-quarter report, which would give back the report's gain. Fourth-quarter 2026 cloud subscription revenue below $143 million, the low end of guidance,…
How this part is made
Claude writes down the test before the outcome is known, so the note can be checked against what happens next. For a note rated High or Medium the price is a close below a stated level; for a note rated Low, which expects the shares to lag, it is a close above one. It tests the note, not the portfolio: the portfolio sells on the business facts it wrote down when it bought, or when that day's research finds the stock fully priced.
-
Both sides, built the same way, with sources and dates.
APPN → Both present, sourced and dated
How this part is made
Claude writes the bear case with the same care as the bull case: sourced, dated points. The evidence against the call is on the page, not just the conclusion.
-
Where the stock trades now: its trend, momentum and the price levels that matter.
APPN → Present: trend, momentum, key price levels
How this part is made
Code computes the price context from daily closes: moving averages, RSI, the 52-week range and key levels. Claude writes the setup from it, so the thesis rests on how the stock trades, not only on the story.
-
The dated events ahead that could move the thesis, usually through the next results report.
APPN → Next dated catalyst · 11 January 2027
How this part is made
Results, product launches and regulatory decisions, each dated, so you know when the question gets answered. Research ages fast, which is why every page shows when it was written.
-
The evidence that would change the rating, stated plainly.
APPN → Present: what would change the rating
How this part is made
Separate from the kill line: the evidence that would move the rating up or down without ending the call. It shows what the research is waiting for.
-
What drives the stock, which decides what evidence the research weighs.
APPN → Second-order beneficiary
How this part is made
One of seven archetypes, from theme leader to event-driven. An event-driven stock is read for its date; a crowd-driven one gets a tight kill line. See the full list.
Archetype taxonomy → -
The note's three-month call against the S&P 500.
APPN → Medium
How this part is made
Three levels. High and Medium say the shares are more likely to lead the S&P 500 over the next three months, High on stronger evidence; Low says they are more likely to lag it. Whether the portfolio owns the stock is a separate decision, made by Claude on its own written case for the company.
Conviction, each note's three-month call against the S&P 500: High and Medium expect the shares to lead it, Low expects them to lag it.
-
The themes the stock belongs to, and how it trades with them.
APPN → AI enterprise software
How this part is made
Themes come first: Claude groups the strongest stocks, and the ones the portfolio holds, into themes from their news and filings, and each stock is read against its group and the market regime, not in isolation. The last section of every research note says which stocks it moves with.
Browse this theme →
Four research blocks, and the filed numbers
Once a note has been re-read after news about the company, its stock page also carries four research blocks. Claude writes them in the same read as the rest of the note, and each is dated by that read.
The business
What the company does and how it makes money.
What is changing, and what follows
Three lines: what is changing now, what follows for the company, and what follows for its customers, suppliers and competitors.
What the price assumes
The expectations already built into the share price, with the valuation they rest on and its date.
Linked companies
Its customers, suppliers and competitors. A company the research covers links to its own page.
Filed numbers
What the company reports to the SEC in its quarterly and annual filings, each figure with its period and filing date, then its valuation at the last close, short interest and insiders' open-market trades in dollars. Code gathers these from the SEC, FINRA and daily prices, with no model involved.
Three rules every research note follows
A kill line
Every thesis states the condition that would prove it wrong, before the market decides.
A date
Every research note shows when it was written. Research ages fast, so the date is always on the page.
Both sides
The bear case gets the same care as the bull case. The evidence against the call is shown, not buried.