Perhaps, if you're looking at a language in its entirety, without regard to usage frequency of particular utterances.
However, most of the exceptions tend to be concentrated in the most frequently-used portions of the language. So, while the statement is technically correct, extensive focus on the rules is not really practical.
Take a random verb in English. It's reasonable to assume the third-person singular form can be constructed by appending an "-s" to it. Similarly, the past-tense forms can be constructed with an "-ed." Most verbs are like that.
Yet this rule isn't of much help with the two most common verbs: "to be" (am/are/is/was/were, not bes/beed), and "to have" (has/had, not haves/haved), as well as dozens of others that also happen to be among the most frequently-used.
Thus, from any practical point of view (such as a language learner's), it's best not to expect any rules to hold in principle, at least not until one's awareness of the exceptions is sufficiently advanced.
> And linguistics as a field is really about the regularity of languages. When exceptions are studied, it is to discover the underlying rules that e.g. cause exceptions to arise or to disappear.
I think this is a very good point. I'm not criticizing the questions as they are, as I said I found them interesting. All I'm saying is that it's good to be aware of the limitations of such an approach. It's best to look at some of the problems as logical puzzles, only wrapped in references to some (more or less) obscure languages, and natural languages should not be expected to follow the principles of logic.
Even if the exceptions outnumber the rules, each rule applies in a much larger number of cases, so each case is more likely to follow the rule rather than being an exception.
And linguistics as a field is really about the regularity of languages. When exceptions are studied, it is to discover the underlying rules that e.g. cause exceptions to arise or to disappear.
As a competition, the IOL additionally has the constraint that it should be solvable in a reasonable timeframe, so rather than making people wade through mountains of incomplete and contradictory data, you get carefully selected problems that require less effort to solve.