Experiment · Testing

Attention ledger

An investigation into the gap between a clean engagement number and the much messier experience of paying attention.

Central question

Can an attention framework remain useful without pretending to be a single definitive score?

Current state

The first version separates attention into three conditions: arrival, meaningful time, and return. It is deliberately incomplete.

The current work is to find signals that do not confuse friction with interest, or a long page with a useful one. A good model should make a better question possible, rather than become another number to optimise in isolation.

Next direction

Compare the model against a small set of product and editorial journeys, then remove any part that only looks elegant on a slide.