Thanks Mark. Do let us know the outcome, because I think a lot of API users have been working on the assumption of unique link-strings.
Now I am wondering how many other pairs of shared links there are. The strings have 162 bits (27 characters each of 6 bits) of entropy, so that is about 5.84*10^48 possible strings. The odds of a random collision are vanishingly low if the strings are pseudo-randomly generated. In any case, they should be generated and then checked against the existing list to avoid that. I am wondering if your algorithm involves some sort of hashing routine?
In this example, the underlying documents show Mr Walsh as a director of The Heritage and Mr Cornelius-Wheeler as a director of 80 Claverton Street. Neither is a director of the other company. Perhaps by coincidence, both were born in 1923 but in different months: Mr Walsh in December and Mr Cornelius-Wheeler in March.
In the search page of your web site, a search for Mr Cornelius-Wheeler's name brings up the top match of Mr Walsh:
Thanks again. Very worrying.