Figure 25.2 The autoresearch loop. A change is proposed, run, and measured against a single
scalar metric; if the number improved the change is kept, otherwise discarded, and the loop turns
again—for long stretches with no one watching. The human sits upstream, defining the objective
and the limits once and then stepping away. Everything hangs on the measurement station (in accent):
the loop is only ever as trustworthy as the metric that gates it.
