not if you incorrectly estimate that the broken sensor has zero variance. In that case the kalman filter will discard all the other information and trust it completely. I think that's the point the poster is making (the article uses a somewhat naive estimate of variance)