Suppose an arbitrary scoring system, bogopoints.
Users submit "+1"s for Wikipedia articles that they feel need checking. Players may select arbitrarily which articles they want to access; the +1 count determines the payout (in bogopoints) of a successful fact-check. A bonus applies for articles that are not in categories or general topics that they frequent.
A fact-check is accomplished by:
1. Reserving an article for 72 hours.
2. Submitting a report, with references, that the article is validated.
3. Passing review by three random other users. Review is also incentivised with bogopoints.
A group of people may collaborate on fact-checking, in which case each person of from the group of N people gets the fraction (n ^ 0.25 / n ^ 1.2) of the bounty (so that the benefit is better than working alone for small groups, but negligible for larger ones.) This curve may be enhanced by reducing the second exponent or enlarging the first in order to accommodate exceptionally long articles which need to be broken up.
Habits on the site are determined in two senses:
1. 'Frequenting' a general area (as mentioned in the third paragraph) is accomplished by the union of all IP addresses the player has used to access the fact-checking site, as well as the user account(s) the player has identified as his or her own. If these addresses are used to edit a Wikipedia article which belongs to a category, then all of the ultimate ancestors of that category are considered frequented. There are many categories for which this would make no sense (e.g. stubs), so fuzzy thresholds and manual pruning will be necessary. Discussion pages and votes for deletion are considered part of an article. Since IP addresses may belong to public computers, a relatively short memory may be required between fact-checking site visit and edit time for IP relevance.
2. Authors are not eligible to review any article they have edited. This definition includes the authors' own declared Wikipedia usernames and their known IP addresses. Discussion pages and votes for deletion are considered part of an article. Sock-puppeting is trivial, but hopefully the peer review stage will catch most such forms of corruption.
All of the above assumes no access to the actual Wikipedia database. If that is not the case, the following improvements are possible:
1. Sock puppets can be traced to users through IP addresses.
2. Viewing habits can be incorporated into the 'frequenting' habit model.
3. Viewing habits and extremely large numbers of users can be used to identify public computers and reduce their relevance time.