When the objective of a researcher is to analyse an unobservable quantity that varies through time and there are multiple valid candidates that assert a capacity to estimate it, a procedure that identifies the bias and variance of such estimators is of utmost importance. In this paper, we put forward the ALARM model, which can delimit the error structure of each estimator while using their respective estimates to propose a new one. Using carefully selected priors, we are able to evaluate our assumptions by adhering to the weak severity principle, which is an essential step since the latent variable can never be observed, so extra care is needed to justify the underlying model. Then, as an application, we analyse five estimators of the GDP gap (Blanchard-Quah SVAR, production function, Areosa method, Hodrick-Prescott and Hamilton filters), showing that the Areosa estimator seems to be the most adequate to explain the brazilian output gap.