Inversion of effect direction in multivariate count model

Hello, I am troubled by a strange, albeit not really significant, inversion of effect direction in a simple model. I am comparing 3 count quantities between years, two upper-bounded with a binomial distr., and another with a Poisson distr. This is the model:

brm(mvbf(
       bf(Amb | trials(Slot_DCA) ~ (1 | Year)) + binomial(),
	   bf(nDCA | trials(Slot_nDCA) ~ (1 | Year)) + binomial(),
	   bf(Rep ~ (1 | Year)) + poison()
    ), data = data %>% filter(Period == Period[1]), # don't mind this
	chains = 0, cores = 8, file = fit_file, refresh = 0, iter = 10000,
	backend = 'cmdstan', prior = c(
	        prior(student_t(3, 0, 2), class = 'Intercept', resp = 'Amb'),
	        prior(student_t(3, 0, 2), class = 'Intercept', resp = 'nDCA'),
	        prior(student_t(3, 0, 1.5), class = 'Intercept', resp = 'Rep'),
	        prior(student_t(3, 0, 2.5), class = 'sd', coef = 'Intercept', group = 'Year', resp = 'Amb'),
	        prior(student_t(3, 0, 2.5), class = 'sd', coef = 'Intercept', group = 'Year', resp = 'nDCA'),
		prior(student_t(3, 0, 2.5), class = 'sd', coef = 'Intercept', group = 'Year', resp = 'Rep')
    ), adapt_delta = .9)

And this is the data:

For a particular period (showed above in the filter) I have that the posterior distr. of Amb in 2020 is lower than in 2019, even if the data is the opposite, that is 10 cases (Amb column) over 25 (Slot_DCA column) in 2019 and 10 over 17 in 2020:

The pattern repeats even refitting the model, so it’s not a random fluctuation.

Could it be the multivariate influence in the outcomes? You can notice a big difference between years for the nDCA quantity, which is strongly decreased in 2020. Could this strong effect drag also the Amb down in 2020?