The designers of static analyses used in industry often try to reduce the number of false positives reported by the analysis through increased engineering effort, user-provided annotations, and/or weaker soundness guarantees. To produce a static analysis with little engineering effort, reduced false positives, and strong soundness guarantees in a principled way, we adapt the Abstracting Gradual Typing'' framework to the abstract-interpretation based program analysis setting. As a case study, we take a simple static dataflow analysis that relies on user-provided annotations and has nullability lattice $N \sqsubset \top$ (where $N$ meansdefinitely not null'' and $\top$ means possibly null'') and extend it by adding $?$ as a third abstract value. The question mark explicitly representsoptimistic uncertainty'' in the analysis itself, supporting a formal soundness property and the ``gradual guarantees'' laid out in the gradual typing literature. To evaluate our gradual null-pointer analysis, we implement it as a Facebook Infer checker and compare it against the existing null checkers in Facebook Infer. A preliminary set of experiments show evidence of reduced false positives. We then generalize this example into a system that gradualizes any dataflow analysis in the same way: augmenting the lattice while retaining the properties that permit it to be used in the standard dataflow analysis fixpoint algorithm.