Introduction Within the typology of word prosody, there is a growing body of evidence for initial three syllable windows in which primary accent must be located on the first three syllables of the prosodic word (Caballero 2011, Kager 2012). In this paper, we argue that Ese'eja (Takanan; Bolivia/Peru) also demonstrates this rare type of window. We report on findings of a verbal accent study (Vuillermet 2012), which systematically documents the position of accent in a large number of inflected verbs, resulting in a corpus of nearly 2,000 verb forms. This corpus demonstrates a three syllable window, and reveals a complex but consistent system varying according to (1) stem syllable count, (2) root transitivity, and (3) the type of suffix added. Accent position cannot be accounted for by reference to purely phonological structure such as syllable count, the quantity of the syllable, the quality of the vowel, or the sonority of the onset. We decompose the Ese'eja metrical window into an interaction of morphologically-assigned accent patterns from various suffixes, using common OT constraints. We classify inflectional suffixes into dominant, recessive, and rightmost-preserving, and tense/mood suffixes in particular to different Suffix Accent Groups (SAGs). Each SAG type triggers a specific cophonology (Inkelas & Zoll 2007). SAGs 1 and 3 assign accent to the ultima of the stem, while SAGs 2 and 4 assign accent to the penult of the stem. Additionally, SAG 1 suffixes have a cophonology ranking which enforces the creation of iambs (RHTYPE=I >> RHTYPE=T), whereas SAGs 2/3/4 enforce trochees. After all inflectional material is added to the stem, iterative footing takes place resulting in rhythmic accent across the word. The leftmost foot in the word receives primary accent, and is realized with primary prominence phonetically. Because iterative footing and rhythmic accent are established before primary accent delegation, we therefore classify Ese'eja accent as a 'count system' with bottom-up parsing, and see it as challenging the Primary Accent First (PAF) model (van der Hulst 1996). We argue that the Ese'eja count system cannot be reduced to a prosodic system with only rhythm plus phrase edge prominence without primary word stress. Further, we contrast our analysis with that of Kager (2012) advocating for ternary feet with weak layering. The main difference between these approaches is that under our analysis, the relative ranking of RHTYPE constraints across co-phonologies accounts for the different patterns, whereas under ternary feet the main parameter involves the ranking of ALIGN-HD-L. Moreover, when morphologically-assigned accent falls outside of the initial three syllable window, we argue that Ese'eja demonstrates a novel type of repair, which we call 'rhythmic repair'. Under this type, when accent is assigned outside of the metrical window, primary prominence appears on a 1 Sincere thanks go first to the Ese'eja people who have engaged with this linguistic research, without whom none of this would be possible. We are...