Do planets remember how they formed?

One of the most directly observable features of a transiting multiplanet system is their size-ordering when ranked in orbital separation. Kepler has revealed a rich diversity of outcomes, from perfectly ordered systems, like Kepler-80, to ostensibly disordered systems, like Kepler-20. Under the hypothesis that systems are born via preferred formation pathways, one might reasonably expect non-random size-orderings reflecting these processes. However, subsequent dynamical evolution, often chaotic and turbulent in nature, may erode this information and so here we ask - do systems remember how they formed? To address this, we devise a model to define the entropy of a planetary system's size-ordering, by first comparing differences between neighbouring planets and then extending to accommodate differences across the chain. We derive closed-form solutions for many of the microstate occupancies and provide public code with look-up tables to compute entropy for up to 10-planet systems. All three proposed entropy definitions exhibit the expected property that their credible interval increases with respect to a proxy for time. We find that the observed Kepler multis display a highly significant deficit in entropy compared to a randomly generated population. Incorporating a filter for systems deemed likely to be dynamically packed, we show that this result is robust against the possibility of missing planets too. Put together, our work establishes that Kepler systems do indeed remember something of their younger years and highlights the value of information theory for exoplanetary science.