It is estimated from how the dispersion of returns scales with the interval you measure over. For a random-walk, variance grows linearly with time, so the standard deviation grows with the square root, giving H = 0.5. Faster growth implies trending, slower implies reversion.
Typical measured values on liquid markets sit between 0.45 and 0.55 and move around a lot depending on window and method. A reading of 0.58 on a two-year daily sample is not strong evidence of anything, and different estimators applied to the same data can disagree by 0.05 or more.
Use it as a rough regime-filter input or a descriptive statistic, not as a standalone signal. Anyone selling a system based on Hurst alone is selling an estimation artefact.
Related: random-walk, mean-reversion-half-life, regime-filter, variance-ratio-test