Inference of a lower bound and strictly positive noise

Why take

t_i = \alpha + \beta \cdot x_i + \epsilon, \quad \epsilon \sim \textrm{lognormal}(\nu, \sigma)

rather than the more traditional log-linear regression,

t_i \sim \textrm{lognormal}(\alpha + \beta \cdot x_i, \sigma).

I would be surprised if BRMS used the OP’s formulation rather than standard loglinear regression.

The Wikipedia isn’t very helpful here, but this is what I’m talking about: Log-linear model - Wikipedia

The previous post on this topic was never answered. :-(