Model-theoretic phonology seeks to formally model natural language as a mathematical system. This approach allows for theoretical arguments about the learnability of natural language patterns, addressing questions of how much input is needed, and whether models can be constructed from unlabeled, mostly positive examples (Heinz et al, 2011). The formal language theory behind model-theoretic phonology has also enabled grammatical inference algorithms, which derive grammars from a sample by making assumptions about the complexity class of the language represented (De la Higuera, 2010). Recent work in this subfield of computational linguistics has revealed that phonotactics, among other natural-language phenomena, can be characterized by subclasses of regular languages. In particular, De Santo & Graf (2019) propose a subregular class of Multiple Input-sensitive Tier-based Strictly Local (MITSL) languages for phonotactics, combining the strengths of sensitivity to local structure when projecting to tiers with the ability to encode multiple tiers of constraints. De Santo & Aksënova (2021) propose an algorithm to learn a MITSL grammar from positive language data. Here, considerations regarding the implementation and use of this MITSL algorithm are presented. The algorithm is evaluated on a series of test languages, including simplified corpus data (Aksënova 2020): the learned grammars are used to generate new members of the learned language, and these are classified as to whether they belong to the target language. Importantly, generated strings that do not belong in the target language are directly traceable to particular subsequences missing in the sample. These results show that the proposed algorithm functions as theoretically predicted, demonstrating the learnability of phonotactic constraints from positive data. The implementation of the learner is furthermore valuable to researchers interested in experimenting with MITSL grammars.