Product Sheets

JAX Envision ™ : A Platform Built for Home Cage Research Figure 2: Envision end-to-end performance across a variety of experimental conditions. Performance is reported by condition category (bars = mean; dots = tested conditions). THE DATA, MACHINE LEARNING, AND VALIDATION FRAMEWORK BEHIND ENVISION™ Figure 1. The importance of compounding error when assessing true home cage algorithmic performance. Behavior Classification Animal ID Tracking BUILDING A DIGITAL CORE OF ANIMAL BIOLOGY Biology is continuous. Most preclinical measurements are not. Episodic, observer- dependent behavioral assays give you snapshot — but what happens between observations stays invisible. That gap isn't just missing data. It's missing context: the difference between a transient response and a trajectory, between adaptation and emerging liability, between a subtle phenotype and the noise floor of your assay. Envision™ closes that gap. By continuously and non-invasively monitoring individual animals in their home cage, Envision generates longitudinal behavioral and physiological records for each animal that both serve as primary endpoints and provide the continuous context that makes every measurement in your study more interpretable. ENVISION DEFINES "VALIDATED" FOR DIGITAL MEASURES Home cage monitoring platforms benchmark classifier accuracy on curated test sets – hand-picked clips where the behavior is often clear and the animal is visible. While these datasets demonstrate algorithm performance under ideal conditions, they do not reflect many of the challenging conditions encountered in real preclinical research studies. All platforms consist of a pipeline of algorithms: tracking, animal ID, and multiple behavioral classifiers working in concert to produce a single data point. Importantly, errors compound across these pipelines, and most published accuracy numbers fail to represent true end-to- end performance a user would experience, see Figure 1. Envision is built differently: performance is evaluated end-to-end, from raw home-cage video through to the individual animal digital measures used in your analysis – validated against every second of annotated study recordings, including occlusions, group interactions, and behavioral transitions – the conditions that challenge an algorithm the most. That foundation is LENS (Longitudinal End-to-end aNnotated Standard); a large, expertly annotated dataset built from real-world studies spanning the environmental conditions encountered in typical experiments: multiple coat colors, light cycles, animal number, bedding, and enrichment types, see Figure 2. Because annotations capture the full spectrum of real behavior—not just ideal examples, Envision algorithms are assessed against all the conditions that matter. The result is a realistic assessment of endpoint performance and greater confidence that the digital measures will remain robust, reproducible, and interpretable when applied to real preclinical studies. And because LENS continues to grow, Envision performance continues to improve over time. KEY BENEFITS Model Version: 2026v5.0 Built for Research • End-to-End validation ensuring optimized performance on all study data • Validated against LENS - real world data • More statistical power to detect effects with digital measures Breadth of Insights • Individual animal tracking reveals the biology between scheduled observations • 10+ measures spanning activity, metabolic, and social behaviors – 24/7 Operational Simplicity • Non-invasive monitoring – no handling, no procedural stress, no observer bias • Data quality from day one – no algorithmic tuning to work in your facility • Scalable to large multi-cohort longitudinal programs 95% 95% 95% 86% End-to-End Accuracy x x =

• Cover