model {
	for (i in 1:N) {
		for (j in 1:M) {
# partition-specific measurement errors
			Y[i,j] ~ dnorm(mu[i,j], tau.e[z[i]])
			mu[i,j] <- b[i] + theta[z[i],j]
		}
# partition-specific variance on random effects 
		b[i] ~ dnorm(0, tau.b[z[i]])
# latent partition labels, no attempt to re-label 
		z[i] ~ dcat(w[])
	}
# AR(1) model for mean trajectories with fixed number of partitions K
	for (k in 1:K) {
		# within trajectory variation
		tau[k] ~ dgamma(alpha, beta)
		# first time points
		theta[k,1] ~ dnorm(0.0, tau.t1)
		for (j in 2:M) {
# variance depends on distance between adjacent time points
			tau.theta[k,j-1] <- tau[k]/(x[j]-x[j-1])
			theta[k,j] ~ dnorm(theta[k,j-1], tau.theta[k,j-1])
		}
# priors on measurement error and random effects
		tau.e[k] ~ dgamma(0.001, 0.001) # Not the brightest
		tau.b[k] ~ dgamma(0.001, 0.001) # priors in the world...
	}
# Dirichlet prior on mixing proportions
	w[1:K] ~ ddirch(delta[])
# Again, gamma prior could be improved
	tau.t1 ~ dgamma(0.001, 0.001)
	alpha <- 0.01
	beta <- 0.01
	for (k in 1:K) {
		delta[k] <- 1.0   
	}
}