In data science health research, secondary use of data has become the norm. Clinical records become training sets for prediction models; archived images become inputs for foundation models; legacy biospecimens are repurposed into renewable cell lines. Governance systems have responded by emphasizing documentation — provenance logs, consent forms, data access agreements — as markers of ethical legitimacy. But documentation of the past does not guarantee ethical adequacy in the present.

In this paper we identified and named a specific governance error: the Continuity Trap, which occurs when a visible signal of continuity in one domain (such as a valid provenance record) is mistakenly treated as sufficient evidence that ethical continuity has been preserved across all domains. We argued that true ethical continuity must be assessed across four distinct dimensions — provenance, semantics, authorization, and relational standing — which can diverge as data are linked, transformed, and redeployed. The policy implication is not blanket re-review of every secondary use, but targeted, trigger-based continuity review focused on the weakest domain in any given research context.