About The Research Note

The Research Note exists because most of what gets written about these subjects is a summary of a summary. A press release becomes a news item, the news item becomes a newsletter paragraph, and by the time it reaches a reader the original number has lost its footnote, its denominator, and sometimes its meaning. This publication tries to do the unglamorous version instead: read the filing, pull the series, do the arithmetic, and show the work.

What gets published

Long-form research notes, published when they are finished rather than on a calendar. In practice they fall into a few recurring shapes:

• Company and industry notes: how a business actually earns its money, what the disclosures do and do not reveal, and where the economics are heading.
• Market and macro notes: a specific question, worked through with data, rather than a running commentary on the tape.
• Technology and AI notes: what a system can demonstrably do, what it costs to run, and which claims about it survive contact with evidence.
• Method notes: how a dataset was built or a model constructed, published so the reasoning can be checked and reused.

There is no newsletter, no membership, and no account to create. Everything is free to read, on the web, at a stable URL. The archive holds every note that has been published, and nothing is quietly removed, corrections are made in the open.

How the research is done

Every note begins at the primary source. If a claim can be traced to a filing, a transcript, a statistical release, a court document, or a peer-reviewed paper, that document is what gets read, and it is what gets cited at the end of the piece. Secondary coverage is used to find the source, never to stand in for it.

The part that matters most is the labelling. A research note is only useful if a reader can tell which sentences are load-bearing evidence and which are the author’s judgement. So each claim is one of four things, and the writing makes clear which:

Verified fact

Something a named primary source states directly: a filing, a transcript, a dataset, a regulator, a paper. It carries a citation, and the citation points at the document, not at a story about the document.

Estimate

A number derived rather than reported. The method, the inputs, and the sensitivity are shown, so you can disagree with the arithmetic instead of taking it on faith.

Assumption

An input that could not be verified and had to be chosen. It is stated before it is used, never after the conclusion it produced.

Opinion

A judgement about what the evidence means. Clearly marked as such, and always separable from the evidence it rests on.

Charts follow the same rule. Every series is labelled, every axis starts where it honestly should, and the underlying figures are given in the note so a chart can never be the only place a number appears. When a piece is wrong, the correction is added to the note itself with the date it was made, rather than edited in silently.

Who writes it

The Research Note is published anonymously, and the work is meant to be judged on its own terms: every claim is sourced, every dataset is shown, and every note can be checked without knowing anything about who wrote it. The research interests are narrow on purpose: capital markets, the economics of artificial intelligence, and the businesses caught between the two. Writing about fewer subjects is the only way to write about them from the source.

Where a note touches a company or an asset in which the author has a position, that position is disclosed in the note itself, at the top, before the argument.

Corrections and contact

Corrections are welcome and are treated seriously: a factual error, a misread table, a broken link, or a source that does not say what a note claims it says. The fastest way to reach the publication is email: hello@example.com. Requests to cover a subject are read; sponsored, syndicated, and pay-to-publish pitches are not accepted, and never will be.