Variance Estimation Protocol

The displayed σ² statistic is produced from rolling log returns extracted from timestamped price observations over a short rolling horizon (30 minutes by default) with near-real-time refresh. Returns are defined as r_t = ln(P_t / P_t-1), then evaluated in a fixed-width window to preserve local regime sensitivity while preventing single-tick dominance.

A robust preprocessing stage removes microstructure artifacts: returns are centered by median, clipped using a scaled median absolute deviation threshold, and then passed into an unbiased sample variance estimator with Bessel correction. This controls transient distortions from low-liquidity prints and isolated spread shocks.

The resulting variance is annualized by observation frequency and mapped into an interpretable display scale. A companion volatility term (σ) and deviation index are derived from the same return stream, creating a coherent set of diagnostics tied to a single stochastic process.

This panel pulls live market snapshots from a production data provider on a rapid poll interval (1 second while active), keeps a short in-memory return buffer, and applies the same estimator pipeline to that live stream. The service boundary remains isolated so providers can be swapped without changing interface semantics, output schema, or downstream rendering components.