This manuscript develops a bounded cross-domain framework for identifying when reliance becomes unsafe before a failure is fully visible. It argues that many harmful events form during a receiver-anchored reliance window: an interval in which a promise, output, label, workflow, diagnosis, protocol, role, or institutional assurance has already begun to guide action, while the capacity, context, protocol, fallback, risk, verification, or correction conditions needed for safe reliance remain insufficiently grounded. The paper does not claim that trust failure, AI overreliance, collaboration failure, medical error, diagnostic error, or usability error are the same phenomenon. Instead, it proposes a disciplined comparison method for tracking a narrower event sequence: reliance uptake, under-grounded support conditions, fallback narrowing, late correction, and downstream burden transfer. It includes admissibility and exclusion criteria, a mapping-status framework for distinguishing direct, functional, partial, and analogical correspondence, nine candidate failure-family lenses, and a five-question reliance safety test for pre-closure review. The contribution is conceptual, classificatory, and methodological rather than an empirical validation study. The framework is offered for later case coding, vignette testing, incident review, ontology refinement, and cross-domain comparison across trust, collaboration, AI-mediated reliance, clinical safety, diagnostic classification, public administration, workplace coordination, and institutional decision systems.