FinRL-Meta is a data-centric library for financial reinforcement learning (FinRL) that addresses the challenges of dynamic datasets, low signal-to-noise ratio, survivorship bias, and model overfitting. It follows the DataOps paradigm to automate the curation of market environments from real-world financial data, including stocks, cryptocurrencies, and ETFs. The library provides a three-layer architecture (data, environment, agent) with a plug-and-play interface for popular DRL algorithms. Key contributions include hundreds of gym-style environments, reproducible benchmarks for trading tasks, cloud-based competition platforms, and educational resources. The work emphasizes data quality and systematic engineering to bridge the simulation-to-reality gap in FinRL, enabling more reliable deployment of DRL agents in real markets.