L.F. Gonzalez Hernandez
Wilson's 'zeroth-level' classifier system, suggested as a minimalist system that would allow a better understanding of the inner workings of classifier systems, has been shown to suffer performance breakdowns in simple Markovian environments due to its inability to support long chains of actions. Cliff and Ross  suggested some possible explanations as to why this may happen. In this paper, their conclusions are re-examined and extended to include covering as the fundamental mechanism that causes the bad performance to be sustained for several trials.
Download compressed postscript file