How a study is made, and what it is not
A study asks the same fifteen questions of every company and answers them only from documents the company filed with the US Securities and Exchange Commission. It shows the sentence behind each answer. It does not tell anyone what to do with their money.
- 12companies studied
- 410quotes that passed the word-for-word check
- 12quotes thrown out by the check
- 46answers marked "could not assess"
Five steps, the same every time
The filings are collected
A script fetches the company's own documents from the SEC: the latest annual report, the one from ten years ago, the proxy statement, results releases, letters to shareholders and filings on officer changes. No model is involved.
A reader pulls quotes
An AI model reads the documents and copies out sentences that bear on each question, word for word, with a note on where it found them.
A script checks every quote
Each quote is searched for in the documents. If it is not there exactly, it is thrown out. So far 12 of 422 were thrown out.
Two readers mark it, blind
Two different AI models mark the fifteen questions from the checked quotes only. Neither sees the other's marks. The study shows the first reader's marks and says where the second landed.
The arithmetic is done by a script
Four questions are sums on ten years of figures the company reported: share count, debt against free cash flow, cash against profit, years of falling revenue and the trend in margins. No judgement goes into those four.
It says when it does not know
Filings do not answer everything. When the evidence for a question is missing or too thin, the question is marked "could not assess" and left out of the score. It is never filled in with a middle mark.
The score is the points earned on the answered questions, out of the points those questions could give. A company needs at least 12 of the 15 answered, and at least two in every area, before its score counts for anything. With fewer, the study says "not enough evidence".
The bar is 90. From 80 to 89 is a near miss. So far 12 companies have been studied and none has cleared the bar. That is a result, and it is published as one.
What this site never does
- No recommendationsNothing here says to buy, hold or sell. A score is how well a company fits a list of questions. It says nothing about whether the shares are cheap or dear.
- No rankings, no best-of listsStudies are listed by result so they are easy to find. The order is not advice.
- No prices of our ownThis site shows no prices of its own, and no study mentions what the shares trade at. The charts are revenue, taken from the company's filings. The Market page has an optional strip of live tickers from TradingView, which stays off until you switch it on.
- No quotes from earnings callsCall transcripts belong to the services that publish them. Every quote on this site comes from a public SEC filing and links to it.
- No hidden interestEach study states whether the author holds shares in the company, on the same page as the study.
Where it can be wrong
The reading and the marking are done by AI models. The check that a quote is real is done by a script, so a quote on this site is in the filing. Whether the quote was read correctly is a judgement, and judgements can be wrong.
Company claims are treated as claims. A company that says it is the leader in its market gets a lower mark than one whose share is confirmed by an outside source.
When a mistake is found, the study is corrected and the correction is noted on the page. If you spot one, write to [email protected].
What is not published
The fifteen questions are public, and so is every quote and every note behind a mark. Three things stay private: the exact description each mark is held against, the filter that decides which companies get studied, and the instructions the readers work from. They are fixed before the filings are read and are the same for every company.