The reuse of retired electric vehicle (EV) batteries in electric grid energy storage emerges as a promising strategy to address environmental concerns and boost economic value. This study concentrates on devising health monitoring algorithms for retired batteries (BMS$_2$) deployed in grid storage applications. Over 15 months of testing, we compile, analyze, and publicly share a dataset of second-life (SL) batteries, implementing a cycling protocol simulating grid energy storage load profiles within a 3 V-4 V voltage window. Four machine learning-based health estimation models, relying on BMS$_2$ features and initial capacity, are developed and compared, with the selected model achieving a Mean Absolute Percentage Error (MAPE) below 2.3% on test data. Additionally, an adaptive online health estimation algorithm is proposed by integrating a clustering-based method, limiting estimation errors during online deployment. These results constitute an initial proof of concept, showcasing the feasibility of repurposing retired batteries for second-life applications. Based on obtained data and representative power demand, these SL batteries exhibit the potential, under specific conditions, for over a decade of grid energy storage use.