How we work
A research desk that reads every filing the same way.
The desk is a set of machine researchers, an audit layer written in code, and one human editorial gate. The researchers do the reading and the judgment. The code does the accounting, the arithmetic and the consistency checks, and refuses to let a note through when they fail. A person decides what is published.
What a note is built from
- The annual report, read in full. The business, the risk factors, management's discussion and the market-risk disclosures, with verbatim quotes that are located in the filing before they are allowed to stand.
- A fact sheet from the filing's own XBRL data. Revenue, cash flow, debt, taxes and share counts with a stated definition for each derived figure. Free cash flow to the firm, net debt and the statutory tax rate are computed one way, for every company.
- Twelve years of history. Growth rates, the worst declines the company has lived through, and its own valuation range, computed from the filings as they stood on the research date.
- A discount rate with an anchor. The risk-free rate on the day, a measured beta, the company's cost of debt read from its interest coverage. The researcher may depart from the anchor, but must say why, and the note prints both.
What the researcher decides, and what it is held to
The researcher chooses the method, the bear, base and bull cases, the multiple, and the call. Each of those choices starts from evidence: a growth band from the company's history and from what the price already implies, a bear case that must at least revisit the worst decline in the record, a multiple checked against the company's own range and its peers. Departures need a stated reason, and the reason is printed. The researcher must also say why the opportunity exists at all, in a form the code can test against data the researcher never saw.
What breaks the case
Every note carries the two or three variables that decide it, the external factors the value is exposed to, and a table of break points: what the base case is worth if the discount rate rises a point, if growth runs five points slower, if starting cash flow is ten percent lower, and the exact rate and growth at which the case equals today's price.
Before anything is published
- Every recorded calculation is reproduced independently, in exact arithmetic, from its bound inputs. If one fails, the note does not qualify.
- The call must agree with the numbers. A long whose base case sits below the price is rejected by the software, not by taste.
- Weaknesses the software can see but not settle, a bear case that never loses money, two methods that disagree, a rate off its anchor, are recorded as cautions and lower the note's rank.
- A person reads it and decides whether members see it. Nothing is published automatically.
Keeping score
Every published call is followed at one week, one, three, six and twelve months against its own price and against the market. Hit rates are kept by horizon, by direction and by the kind of caution a note carried, and they feed back into how notes are ranked. We would rather show a mediocre record honestly than a good one selectively.
What we are not
We are not an investment adviser and the notes are not advice to anyone in particular. They are research, produced mostly by machines, checked by code, and published by a person. The desk runs a paper book on its own ideas and says so. Members agree not to redistribute what they receive.