But humans (and other entities) should not be represented by strings in the system, but by items.
I wonder whether this would not be too inflexible. It would burden the use of wikidata with the responsibility to determine entity-identity in all cases where only a name-string is known.
In the example of the mayor: Assume that the new mayor of a city is named "John Smith". Wikidata already has 500 items for persons named John Smith. The Wikipedia-Wikidata editor must now determine whether it is good practice to simply create wikidata-item 501, not knowing whether it is one of these or not.
I fear that the practice is even more problematic in the reverse case. If in a large percentage of cases there is little doubt about identify, this could lead to the practice of always connecting to a wikidata-item for a person, should there be a person of this name. Henceforth, Wikidata would claim that the mayor of Erewhon previously was councilor in Owd-Negrin, even if there is only a chance identity of a name. Wikipedia disambiguation pages know how many homonymic highly notable persons exist - Wikidata will deal with the non- or less-notable ones as well.
A well known example is that it is not a good idea for scientific reference management to treat authors as person entities, since the "reverse engineering" of author identity from the n:m relation between person and name-string is normally not feasible.
I would prefer if the decision whether entity-identity is known or whether only a name-string or other label is known, should be left to the Wikidata editor community, and not prescribed by the software.
Gregor