The Indus Valley script resists phonetic decipherment because it does not encode phonetic language. This monograph proposes that sustained administrative and trade interaction during the Harappan period (c. 2600-1900 BCE) created a Phonetic Constraint Field—a probabilistic phonotactic environment that shaped spoken languages without standardizing lexicon or grammar. The Harappan Phonetic Constraint Field (HPCF) exerted selection pressure favoring CV/CVC syllables, retroflex-friendly articulation, and cluster-breaking through vowel epenthesis. We operationalize this model through computational constraint filters applied to lexical domains (numerals, pronouns, body parts, motion verbs, material culture terms). The model predicts structural convergence without lexical inheritance—precisely what archaeological linguistics observes in post-Harappan South Asia. Dravidian, early Indo-Aryan, and regional languages show phonotactic commonalities exceeding chance predictions, despite divergent lexical inventories. The HPCF framework explains three phenomena resistant to traditional approaches: (1) why rebus-based decipherment fails across semantic domains, (2) why post-collapse languages share sound patterns but not vocabularies, (3) why the script disappears with urbanism rather than persisting as writing. This study is falsifiable: new linguistic data showing non-convergence, discovery of phonetic encoding, or failure of computational tools to match empirical patterns would refute the model. We do not reconstruct Harappan speech, translate seals, or claim decipherment. We propose a mechanism explaining why decipherment fails—and what the failure teaches us about pre-modern language contact.