Insights

Insights & Field Notes.

Written analysis of merchant-observable agent traffic — what request logs actually show, what they cannot show, and how to tell the difference.

What this is

A small, slow series. Each entry carries a named author, a publication date, and a visible source list, and every factual claim in it is attributable to something a reader can check.

Entries describe merchant-observable evidence only: request-level signals recorded on surfaces the merchant controls. Cartograph does not claim to know what an external AI model concluded, preferred, recommended, or decided off-property, and absence of agent evidence does not prove human traffic.

What you will not find here

  • No benchmark numbers or traffic volumes without a linked source.
  • No inferences about model behaviour observed outside merchant surfaces.
  • No customer data, and no aggregate figures drawn from private deployments.
  • No placeholder or preview entries standing in for work that does not exist.

Read the evidence framework instead

Until the first entry publishes, the reasoning behind Cartograph's approach is already written up across these pages.