Microtubules (MTs) are dynamic polymers with critical roles in processes ranging from membrane transport to chromosome separation. Central to MT function is dynamic instability (DI), a behavior typically assumed to consist of growth and shortening, with sharp transitions in between. However, this two-state assumption disregards details in MT behavior that are evident in high-resolution data. For example, MTs exhibit growth rate variability, and pinpointing where transitions begin can be difficult when viewed at high spatiotemporal resolution. These observations suggest that MT behavior is more complicated than implied by standard quantification methods. To address these problems, we developed STADIA (Statistical Tool for Automated Dynamic Instability Analysis). STADIA's methods are rooted in machine learning to objectively analyze and quantify macro-level DI behaviors exhibited by MTs. Applying STADIA to MT length-history data revealed a transient, intermediate phase that we term 'stutter', during which the rate of MT length change is smaller in magnitude than growth or shortening phases. Significantly, most catastrophe events in both simulations and experiments are preceded by stutters, suggesting that this newly recognized phase is mechanistically involved in catastrophes. Consistent with this idea, a MT anti-catastrophe factor (CLASP2γ) increases the likelihood of growth following a stutter phase in experiments. We conclude that STADIA enables unbiased identification of DI phases including stutters, producing more complete and accurate DI measurements than possible with classical analysis methods. Identifying stutters as a distinct and quantifiable phase provides a new target for mechanistic studies regarding DI phase transitions and their regulation by MT binding proteins. 5The advent of high-resolution data acquisition has revealed an additional problem with standard DI analysis: it can be difficult to determine with reasonable precision where transitions between phases begin and end (Figure 1 C,E). This observation leaves researchers to make subjective judgments or to use 'in-house' software with non-adaptive criteria to identify the points where phase transitions occur (e.g., (Yenjerla, Lopus, and Wilson 2010; Goodson and Jonasson 2018; Zanic 2016)). To illustrate this problem, consider the zoomed-out length-history plots that are typically used for DI analysis (Figure 1 B,D). Examination of these plots can make the task of determining when transitions occur look trivial. However, the zoomed-in views made possible by high-resolution data acquisition demonstrate that any software or method using the aforementioned two-state behavior framework will have difficulty categorizing ambiguous behavior that often occurs between growth and shortening phases (Figure 1 C,E).Taken together, these issues indicate that there is significant need for an improved method of characterizing MT length-history data that removes the two-state behavior assumption and allows for unbiased, objective quantification of MT behavior a...