10 estimate the total number of unique ideas that can be generated. Kornish & Ulrich 2009 propose a strategy similar to the “Mark and recapture” approach in ecology (catching a fish, marking it, and throwing it back in the pond). One calculates the number of ideas (T) for a given space by relying on the total number of ideas (N) and the number of unique ideas (U). The equation is derived from models that describe how the probability of new discoveries (unique findings) decreases as more samples are taken from a finite population. The full calculation can be found in Appendix F. To calculate U, each new idea is compared to all previous ideas using cosine similarity. This list of results is then compared to our threshold of 0.8 to find identical ideas. If one of the comparison cosine similarities is greater or equal to 0.8, we consider the new idea identical to an existing idea and hence increase U. N is given by the total number of ideas we have generated. As noted before, cosine similarity is an imperfect measure of similarity, so different methods and thresholds will yield different results just as discussed in Kornish & Ulrich 2009. In addition, it is important to highlight that we cannot estimate the total number of ideas in a space, but only the possible number of unique ideas that can be found with a particular technique. Nevertheless, it is a helpful measure to compare different techniques and their theoretical limits. Speed of Exhaustion Lastly, we consider the speed of exhaustion which describes the rate a strategy depletes its reservoir of unique ideas. If we again assume that there exists a finite but large number of ideas for a specific domain, our first few picks are likely to be very different from each other (low similarity). With each new pick, the likelihood of encountering a similar idea increases as discussed above. We compare each new idea to all previous ideas using cosine similarity. We then look for the most similar idea in the existing set, i.e., the max cosine similarity score from all our comparisons. To prevent outliers from having an outsized effect, we apply exponential smoothing (alpha = 0.5). This gives us the cosine similarity on the y-axis for each new idea in relation to all previous ones. Technical Set-Up