At q32, after 32 passages of development, we copied each organism in a new cohort into eight branches and gave each branch a different but equally strong local push. Four passages later, a score calibrated on earlier organisms selected the siblings at the two ends of that scale, before any of their long futures had been run; both then received the same continuation and eight later actions. Across the 30 ancestors for which both branches completed, the higher-scoring sibling occupied a narrower microscopic response repertoire, while the experiment’s main comparison of transition timing still included zero.
The images below come from the pair whose field repertoire contracted most strongly, which makes the difference easy to see and also makes this an example chosen after the analysis; it illustrates what the comparison measures, while the result is the average across all 30 paired ancestors.
Within this pair, the score separated the siblings by 0.432, mean pairwise final-field breadth moved from 0.234 in the lower-scoring branch to 0.178 in the higher-scoring branch, and the timing range moved from 5 to 4 steps.
At q32—after 32 passages of development—we copied the same ancestor into eight branches and applied a different first action to each at q33. At q36, a previously calibrated commitment coordinate, a score built from two earlier information measurements, selected the siblings at the two ends of its range before any long continuation had been run; at q37 both received one unforced continuation and the same eight later actions, giving nine final fields per sibling to compare at q900.
Each value subtracts the lower-scoring sibling from the higher-scoring one, so a negative field value means that the higher sibling reached a tighter set of q900 fields. The interval comes from 5,000 bootstrap resamples of the 30 ancestors, which kept the two siblings from each ancestor together.
Not every organism responded alike: 18 higher-scoring siblings narrowed and 12 widened, yet the mean stayed below zero whenever we omitted one ancestor at a time; seven of the eight morphology-group means were also negative, although each group contains only three or four ancestors.
of the higher-scoring siblings ended with a narrower field repertoire; the other 12 widened.
of the morphology groups had a mean below zero; after omitting one ancestor at a time, the overall means ranged from -0.0089 to -0.0061.
Transition-timing range was the experiment’s predeclared primary comparison, while final-field breadth was one of its named secondary measurements. Timing narrowed by an average of -2.57 steps, but the 95% interval ran from -6.67 to +1.00; with that interval, the timing data could reflect a modest contraction, or no change at all.
of the paired ancestors had a narrower timing range in the higher-scoring sibling, while 6 were equal and 10 widened.
pairs transitioned under every arm in both siblings, so the intervention did not change event incidence itself.
Seven group means fell below zero and one rose above it, but each group contains only three or four ancestors, so this view shows where the overall average came from without making any one small group a result of its own.
Within this pair, the score separated the siblings by 0.545, mean pairwise final-field breadth moved from 0.472 in the lower-scoring branch to 0.449 in the higher-scoring branch, and the timing range moved from 19 to 35 steps.
The two descendants began from the same ancestor and were separated by different selected first actions before their long futures were run, so the coordinate-guided policy caused a change in descendant repertoire; it does not yet show that the coordinate itself caused the contraction, because choosing the two ends of that score also chose different first actions. These runs follow a single developmental transition in each arm, and therefore do not test autonomous replication.
Across this cohort, the coordinate-guided first-action policy produced higher-scoring siblings whose later responses to the same action family occupied a smaller mean field repertoire.
The next run should compare one, two, and four coordinate-directed pushes, then replay the same later actions after the policy is released; this would show whether the contraction grows into a stable developmental change or fades with the steering.