We present results from a large-scale online survey (N = 4545) and an in-depth follow-up study (N = 48) showing unexpected variation patterns in the case marking of theme subjects in Icelandic. In previous accounts (Jónsson 2003), quirky oblique theme subjects, in opposition to experiencers, were thought to obligatorily pattern with the structural nominative case instead of a possibly inherent dative case (Nominative Substitution rather than Dative Substitution). Contrary to this, we show that the productivity of dative case marking with themes is non-negligible and possibly increasing, pointing towards a much more diffuse directionality of the change. We argue that this is not unexpected if we assume a theory where semantic roles are probabilistically mapped onto case and productivity is driven by type frequency, as it is under the Tolerance Principle (Yang 2016). We put this to the test by combining a verb knowledge task with a case selection task to compute productivity patterns, and individual type frequency counts, predicting the presence of dative productivity correctly in 39 out of 48 participants by using the Tolerance Principle. Although our results disconfirm a categorical impossibility of theme subjects receiving dative rather than nominative, the previously attested patterns still hold. Experiencers appear significantly more frequently with non-nominative substitution than themes do, originally accusative subjects have the highest rate of substitution and verbs which are generally more known/frequent are less likely to show subject case substitution. These patterns are reflected in attested examples of non-nominative subjects with novel verbs.