Papers

How we work, written from our own projects.

Each paper takes work we did on our own sites and publishes the reasoning, the checks and the failures. Every paper has a section on what failed, because a setup that has only ever worked has not been run.

Paper 01 · 28 October 2026

One Conductor, Many Instruments

How we run one orchestrating model, outside models commissioned per question, and a separate checker ... with the owner holding every key

The seats, the fence between them, and a worked example from our own sites: three outside models scored our home pages, a separate checker tested their claims live, and three of six did not hold up and two held only in part or with different numbers. Plus what failed, and what is still open.

Read the paper →

What these papers are

They describe our own work, not client deployments. Where we measured something we say how. Where we did not measure it we say that, instead of implying a result.

Paul Ann Media publishes the other side ... the same work from the brand side, with the founder.