This paper extends the rough volatility hypothesis, originally demonstrated using realized volatility, to range-based volatility estimators such as Garman-Klass and Parkinson. By analyzing a diverse dataset including non-standard assets (e.g., S&P 400, IBEX 35, individual stocks), the author confirms that log-volatility exhibits scaling properties consistent with fractional Brownian motion with a Hurst exponent significantly below 0.5, often below 0.1. This finding addresses concerns that rough volatility might be an artifact of microstructure noise in high-frequency data, as range-based estimators are robust to such noise. The study validates the rough fractional stochastic volatility (RFSV) model and demonstrates its superior forecasting performance compared to traditional models like AR, HAR, and GARCH. The results provide strong evidence against the fractional stochastic volatility (FSV) model, which assumes long memory with Hurst exponent > 0.5, and support the intrinsic rough nature of volatility across various asset classes.