A consortium of 15 institutions, including Texas A&M University, Google DeepMind, CMU, Stanford, Microsoft, and NVIDIA, released TouchScale, a 500-hour human vision-tactile dataset built with unified sensors and collection pipelines, containing about 87,000 interaction recordings and over 1,500 objects. Using it for mid-training improved the average success rate on four real-robot contact tasks from 22.5% to 57.5%, showing scaling benefits.