Independent Analysis | Three Douglas, LLC
Research Process

How we write, and how we keep score.

The method behind every Three Douglas note, and the record we keep on ourselves.

Most research asks to be trusted. Ours asks to be checked. Every thesis we publish is built against the same standard, and the standard is public, so you can hold any note we write against it.

The method

  1. A named framework.Constraint migration. The Token Tollbooth. The Three Waves. A thesis that cannot be named cannot be tracked, argued with, or falsified.
  2. The conclusion and the tension, up front.The position and the strongest fact against it appear in the first paragraphs, together. Burying the conclusion is retail writing; burying the tension is sell-side writing. We do neither.
  3. Dated data anchors.Every load-bearing claim carries a specific number, a date, and a source. A thesis paragraph without a number in it is usually a slogan.
  4. The steelmanned bear.The opposing case is presented at full strength before it is answered. If the bear case reads weak, we have not understood it yet.
  5. Handicapped risks.Risks carry a probability or severity judgment and a mitigant. A bullet list of everything that could go wrong is liability management, not analysis.
  6. Break conditions, in advance.Every framework states what evidence would prove it wrong, before the fact, and we track those conditions quarterly. A framework without break conditions is a religion.
  7. Track-record honesty.We keep our misses on the record next to our hits. Credibility compounds from admitted misses faster than from claimed hits.
  8. Scenarios that tie out.Bull, base, and bear cases carry explicit probabilities, and the probability-weighted value reconciles to the conclusion. If the scenarios do not tie to the number, the number is decoration.
  9. Valuation with the levers disclosed.Multiple methods, with each method's weakness stated plainly, including which assumption does the work. Showing the lever is more credible than hiding it.
"Frameworks that cannot be falsified are marketing, not research."

Kept on the record

Three examples of the scoreboard we keep on ourselves, because a method note that only lists wins is itself a violation of the method:

We missed memory. One of the larger re-rates of this cycle happened in a sector we surveyed and passed on. The lesson is codified: the window between first evidence and full crowding is compressing, and surveys must convert to conclusions faster or say why not.
We overstated a claim in public. An early draft of a public page used the word "only" where "only" was not true. It was corrected the same day, with sources, and the corrected argument was stronger. Absolute words now get a dedicated fact-check pass before anything ships.
Our early models took shortcuts. An early long-dated valuation was a terminal multiple in a trench coat: no interim cash burn, no funding cost. The rebuild cut the valuation and improved its quality. If the model does not pay for the construction and the dilution, it is not a model.

Where the method shows up

You can see the standard applied in our public notes: our August 30 NVIDIA note publishes five explicit break conditions on our own constructive view, and our July 27 commentary steelmans the vendor-financing bear case before answering it. Members receive the full stream this method produces: coverage notes, quarterly outlooks, and thematic frameworks, as they publish.

Membership is $99.95 per year: the full coverage stream, delivered as it publishes.

No payment today; founding rate locked. First member mailing: mid-September 2026. Membership details