Cardiopulmonary resuscitation (CPR), a basic life-saving skill, requires a combination of procedural and declarativeknowledge. CPR proficiency was assessed and re-trained to criterion across four sessions (spaced weeks to months apart).In addition, three laboratory tasks were administered: continuation tapping, paired-associate learning, and Raven ma-trices. These served as proxies for procedural learning, declarative learning, and general cognitive ability, respectively.Even though a computational model (Predictive Performance Equation, Walsh et al., 2018) predicted long-term CPR per-formance, none of the lab tasks correlated with any aspect of CPR performance (initial performance, (re-)learning, orretention of CPR; see https://osf.io/m8bxe/ for details). These results highlight the challenges faced when translating labresults into real-world domains and can serve as a benchmark for applying computational models to real-life learning andforgetting.