Experiments that seem most informative can fail to resolve critical structural questions.

Scientists usually pick an experiment by estimating which one gives the most overall information. However, some problems suffer from partial identifiability, meaning background noise or nuisance factors prevent researchers from figuring out the true underlying structure from an observation. Standard information measures can fail here. To solve this, researchers introduced Resolution-Aware Experimental Design, or RAED. Instead of chasing raw information, RAED chooses the experiment that produces the smallest set of possible true answers while keeping false exclusions under control. It works by directly narrowing down valid structural candidates without mistakenly ruling out the correct one. The authors proved mathematically that standard methods can pick an experiment with arbitrarily worse structural resolution. In tests on subsurface-flow and chemical benchmarks, RAED selected better experiments and traded risky false exclusions for honest ambiguity.

Why it matters

Scientists designing experiments often choose tests expected to yield the most insight. But in systems with hidden variables, some informative-looking experiments leave key questions unanswered. RAED helps select experiments that genuinely narrow down possibilities, improving reliability in fields like environmental science.

Questions answered