Applied to statistical physics models, the random cost algorithm enforces a
Random Walk (RW) in energy (or possibly other thermodynamic quantities). T
he dynamics of this procedure is distinct from fixed weight updates. The pr
obability for a configuration to be sampled depends on a number of unusual
quantities, which are explained in this paper. This has been overlooked in
recent literature, where the method is advertised for the calculation of ca
nonical expectation values. We illustrate these points for the 2d Ising mod
el. In addition, we prove a previously conjectured equation which relates m
icrocanonical expectation values to the spectral density.