I don't work for Splunk, so can't answer this with 100% certainty, but the fact that the first 15 digits are fine and the last 9 are random suggests that internally Splunk is storing numeric field values as IEEE double precision floating point values, which have 15 digits of accuracy.
Splunk doesn't corrupt text fields that are longer than 15 characters, so one workaround would be to put a letter at the beginning of your 1s and 0s so that the overall field is text rather than numeric. Presumably you're testing individual bits to see if they're 0 or 1 using substr() or similar, so you could still do this if there was a letter at the beginning of the field.
