The "football is random" thesis is built on data overwhelmingly drawn from teams that play high-entropy chains. When the chain is constructed deliberately on top of a developed technical substrate, its entropy drops and its expected outcomes rise. This experiment lays five constructed chains side by side and reports the entropy and expected goals of each, so the trend is visible directly. The chains are illustrative parameterizations consistent with the literature on possession-quality differences by level, not calibrated estimates from any specific competition.
As playing level rises, the mean row entropy of the Markov chain decreases monotonically while the expected goals per possession rises. The same teams the public discourse points to when claiming football is random are clustered at the high-entropy end. The selection effect is the explanation. Strip out the lower-level chains and the variance picture changes entirely.
Entropy and expected goals move in opposite directions as playing level rises. The recreational chain has roughly twice the row entropy of the world-class chain and roughly one-fifth the expected goals per possession. The relationship is not subtle; the substrate determines the chain, the chain determines the outcome distribution, and the gap between levels is mostly a gap in substrate.
The clear implication for the analytics literature is that any cross-level conclusion about football randomness needs to disaggregate by substrate level before it generalizes. The implication for development is that the developmental yield curve from substrate work is steep at the low end and flattens at the top, which is why the same coaching content produces dramatically different chain entropy at different developmental stages.