I've actually run into this in the wild, with regards to sales forecasting. A program we were using returned zero if the error bars on a forecast were over 100%. For example, selling somewhere between 1 and 7 units, but averaging 3.
replyReturning 3 was "wrong", but infinitely more correct than retuning 0.
The article server is offline, but I assume they found out that prolog rule evaluation depends on the order the rules are presented in the program.
replyIf so, the language they thought they were using (and that they should actually use) is datalog, not prolog.
Datalog has declarative semantics: All facts that are derivable from the base database and the rules will be derived by the interpreter, and it will not add extra hallucinated facts. If that's not true, it's a bug in the runtime, not in the language.
www.metalevel.at is run by Prolog legend Markus Triska, author of CLP(FD)/CLP(Z).
replySo it's not that they "discovered" anything about Prolog; they already knew the language inside out.
This article explains how to appropriately use Prolog declaratively and with full generality.
iirc, shor's algorithm for factoring relies on this.
replySometimes the Biorhythm program on my Apple ][ failed to produce correct answers. But it sure was great for impressing cool hippie chicks.
reply