The Pyre sequence ends in a validation plan with a horizon, and the horizon is doing more work than it looks like it is doing.
A change introduced without a stated review date has an unlimited grace period. If it is working, nobody checks, because things appear fine. If it is not working, nobody checks either, because the cost is diffuse and the person who introduced it has moved on to the next thing. Six months later the change is simply how the business operates, and its effect has never been established in either direction.
What a horizon commits you to
The plan states, before anything is built, which number is expected to move, by roughly how much, by when, and what will be done if it does not. The last clause is the one that gets skipped and the one that matters, because a test without a stated response to failure is not a test. It is an announcement.
Horizons differ by surface. A change to how proposals are written can be read in a handful of weeks, because the volume of proposals is high and the loop is short. A repositioning cannot be read in a quarter, because the buyers who will encounter it have not started looking yet. Setting the second one on the first one's clock produces a false negative and the abandonment of something that was working.
When a horizon passes and the number has not moved, that gets recorded as evidence. The hypothesis was wrong, or the build did not match the design, or the horizon was too short. Each of those is a different correction, and naming which one applies is the difference between learning and rotating.