Looks reasonable, I would perhaps use a lognormal likelihood directly to avoid converting the censoring bounds.