Tianye Ma¹ (), Inik Kim¹, Daniel Punzalan¹, Weiwei Zhang¹; ¹University of California, Riverside

The discrete-capacity account of short-term memory (STM) posits a fundamental distinction between stored items and "objective guessing" (zero information). In contrast, continuous-resource models, particularly the Target Confusability Competition (TCC) model, reject discrete limits, arguing that "guessing" is merely an artifact of low signal-to-noise ratios scaled by psychophysical similarity. TCC's superior fit in statistical model comparisons has been widely cited as evidence against discrete capacity. Here, we demonstrate that such conclusions are unwarranted. First, for simulations using a generative process model (combining binding pool and population coding mechanisms) with known discrete limits, TCC consistently outperformed Mixture models, failing to recover the ground truth of capacity bottlenecks. Second, in an empirical experiment, we manipulated the probability of target absence (20% vs. 50%) to establish a ground-truth lower bound for guessing. TCC continued to outperform guess-explicit models (Mixture, TCC-g) even when objective guessing was forced. Critically, TCC-g failed to accurately recover the known guess rates, frequently estimating zero guessing despite 20%/50% target absence. Second, TCC’s strength parameter (d') conflates representational strength with noise. In line with this parameter conflation, an individual-difference analyses reveal that the variance explained by TCC d’ can be partitioned into variance stemming from the capacity limit and the precision limit of STM across individuals. Thus, TCC d’ acts as a composite measure of both quantity (capacity) and quality (precision), rather than a pure index of memory strength. Together these findings indicate that standard statistical model comparisons are insensitive to the structural differences between these theories, rendering the rejection of objective guessing premature.