How should I handle measurement errors in a timeseries database?

I have a table used to record measurements sampled at regular intervals on different sensors. Each row records the time, the identifier of the quantity being measured, and the value itself. Now and again measurement errors occur and garbage is being recorded in the value field. How should I deal with these errors: 1. Delete the offending rows entirely, losing the information that there had been an error; 1. Keep the row as is, and asking of client code to deal with the errors; 1. Replace the values with NULL, losing the original erroneous value? Or is there another option that I have not considered?