The Answer Line

On location, 2026

Screens, at the time of writing

Detail, 2026

01

This does one job. It takes a task that comes back often enough to be worth writing down, and it does that task the same way every time, so the result can be checked rather than taken on trust. Everything else it might have done was left out.

It is used in real work rather than kept as a demonstration. That is the only test it has to pass: it gets reached for before anything else is reached for, and it would be missed within a day if it stopped working.

What it needs to run, what it will not do, and where it keeps anything it has to remember are all stated further down rather than left to be found out. A tool that hides its edges is a tool nobody else can pick up.

In the field, 2026

02

The shape is deliberately ordinary. It reads from one place, writes to one place, and keeps nothing in between that could not be rebuilt from what it read. Anything that has to survive a restart is written down somewhere it can be looked at.

There is no hidden step. Every action it takes is one a person could take by hand, in the same order, for the same result, which is what makes it possible to check the output against the input rather than against a promise.

Editor note: both stops on this page are placeholder. They describe the shape a project entry takes rather than this project, and need replacing with the real account of what it is and how it works.

A readiness assessment for the AI answer era. It scores a site from 0 to 100 across two halves, Reach and Interpret, and positions the result against the Answer Line, the threshold at which a site is answer-ready.

Rather than inferring behaviour from documentation, it probes the site directly: server-level bot access across the AI access matrix, robots and llms files read first-hand, structured data and article bodies fetched rather than assumed.

What comes back is a score with the diagnostic behind it and a plan ordered so the cheapest meaningful fix comes first.