The bilberry-vs.-cowberry fine-scale dynamics has been modelled by a discrete Markov chain of transitions among the four observable statuses of permanent sample plots: bilberry (Vaccinium myrtillus) alone, cowberry (V. vitis-idaea) alone, both species, and species-free (Logofet and Maslov, 2019). Six successive examinations of spp. presence/absence on 20 × 20 cm quadrats located along permanent transects in a Scots pine boreal forest were initiated in 1980, 26 years after a forest fire in 1954, and repeated in every 5 years, thus scoping the 25 years of post-fire succession (Maslov, 1989). Those data have provided the exact calibration of 5 one-step transition matrices forming a nonautonomous Markov-chain model. Its basic prediction (forward forecast) has been obtained as the limiting distribution of states generated by the geometric average of transition matrices featured increase in the share species coexistence.Presented in this paper, the alternate modes of prediction are the validation forecast and the backward prediction. The former is implemented as the geometric average of the first 4 one-step matrices and its dominant eigenvector as the limiting distribution of quadrate states. It appears to hit a 5% vicinity of the distribution observed in 2005, thus certifying a success of model validation.As regards the backward prediction, I propose a technique that realizes the idea of backward prediction by means of reversing the time dimension in the observation data. The ensuing formula for the average backward transition matrix looks somewhat different from that for the forward one but enables calculating the matrix as reliable as the forward formula does. Although the 1955 distribution back-predicted from the observed 1980 one cannot be quantitatively interpreted as a community that might form just 1 year after the fire, it does qualitatively confirm (together with the backward limiting distribution) the directionality revealed earlier in the forward dynamics.