Hi, if you’re comfortable working with Stan directly, the implementation of a measurement error model for p_{prop} would be very straightforward. Given that you know the number of samples N and the observed proportion p_{prop} positive, the “true” proportion positive p_{true} can be assumed to follow a beta distribution with shape parameters \alpha = p_{prop} * N +1 and \beta = N - \alpha +1. This approach assumes a uniform prior for the unknown parameter p_{true}. You can then use p_{true} as a predictor of your outcome of interest.
I don’t know whether this is possible to implement via brms.
Edits: formatting stuff
P.s.: i see that the use of a beta distribution was already suggested. There is no need though to put bounds very close to zero and one. You should be fine putting them at zero and one directly