Research on technical debt in open-source software predominantly examines accumulation, how much debt exists and how long remediation is deferred, implicitly assuming that debt is observable. We argue that observability is itself the prior variable: debt that contributors cannot detect cannot be addressed, regardless of its severity. We present a multi-case empirical study of three latent debt instances discovered and retired in Apache Fineract, a production core banking platform deployed by microfinance institutions worldwide, during a four-month period of active contribution. Each case exhibits a distinct visibility signature that determined both its persistence (four to eleven years) and its detection mechanism. Case A (visible but unowned): security debt documented in public CVEs and wiki advisories for years, yet unaddressed due to the absence of a responsible steward, a condition we term the maintainer gap. Case B (invisible at the execution surface): a REST endpoint fully scaffolded with database columns, permission records, and API routing, yet silently rejecting every invocation since August 2015. Case C (economically tolerated): the same loop-invariant query amplification anti-pattern recurring independently across four organizationally distinct subsystems. Cross-case analysis yields a debt visibility taxonomy and three distinct retirement lifecycles: Warn, Disable, Delete; Discover, Delete; and Detect, Fix. Our findings suggest that debt longevity is governed not solely by remediation cost but also by visibility conditions. We therefore propose debt visibility as a complementary analytical dimension for understanding why technical debt persists and how it is ultimately retired.