Evidence-Obligation Database Theory: What Must Be Learned to Resolve an Unknown Query

Two databases may contain exactly the same information today yet differ fundamentally in what they can know tomorrow. An unresolved query may remain unanswered because the necessary evidence has not yet been acquired, or because no admissible future evidence can resolve the ambiguity at all. This distinction is not captured by current uncertainty alone. We develop Evidence-Obligation Database Theory (EOD), a finite semantics that represents both acquired evidence and the capability to acquire further discriminating evidence. Instead of returning only an answer or uncertainty, an EOD query can return an answer certificate, a minimum evidence obligation specifying what must be acquired, or an impossibility certificate showing why resolution cannot be achieved. We prove that identical current query views can have different future answerability; minimum fixed-evidence obligation is NP-complete; fixed obligations correspond exactly to weighted hypergraph transversals; coarse resolution summaries are insufficient for composition; and signed evidence-incidence annotations enable exact Boolean composition. We further establish a cost-preserving correspondence with conditional sensing, connecting EOD to established information-acquisition theory. Reproducible finite benchmarks and two credit-risk studies illustrate that predictive confidence or apparent saturation need not imply evidence maturity. EOD thereby provides a database-native semantics for unresolved queries as obligations to acquire discriminating evidence.