You will take this Bayes net and use MCMC to approximate Pr(B|C=T). In particular, you will compute, for each node except C, its conditional probability given its Markov blanket. Each run of MCMC will begin with some randomly chosen initial assignment to the variables (with C=T). One iteration within that run consists of iterating through the non-evidence variables (all but C), updating each variable one at a time, using a biased random number generator biased according to the row of your conditional probability table corresponding to the current values of the other variables in its Markov blanket. This will generate 4 new instances.
You will do 5 runs. For each run, you will collect successive approximations of the probability Pr(B|C=T). Every 40 instances, up to 400, record the ratio of instances with B=T to the total number of instances. These are your successive approximations. Graph them. You should either have 5 graphs, or one graph that shows all 5 runs. (Most of your runs should converge to similar values. If not, do more and/or longer runs.)