For this purpose, I think it would be nice to access the raw data, to see any errors that would be otherwise masked. As someone in the comments suggested, one might compare number of corrected errors in 1, 2, 5 years and compare to the number of redundant bits stored to estimate the expected longevity of the medium