read_tsv will read some number of lines to guess the column types, it will then read the rest of the lines, and will tell you about problems where new data is not compatible with the rows used for guessing.
In this case X14 seemed to be logical (really it was all NA's until the first value, which is 189861 i.e. a number which appears on row 11910.
This implies that if read_tsv involved row 11910 in its column type guesstimation, it would guess better the types needed.
This seems to be correct