It seems a bit unwise that the records get automatically merged.

I wonder if anyone did any statistics on the frequency of record collisions between two different passengers whose identities randomly coincide. You can imagine among certain common names, particularly in certain places in the world, it could be relatively common. I suppose the difficulty of trying to get her these statistics would be determining what was a collision (of two different people) and what was a valid merge (probably a rare event anyway). But I think you could get a fair estimate by merging across ignoring one variable (such as date of birth) and then seeing how many of those merges would have failed if the DoB had been taken into account (adjusting for the a priori probability of two DoBs randomly coinciding), and then repeating over other variables, to get an estimate of the relative contribution of uncertainty from each field.

Actually sounds like quite a fun project....