Hurst-Aware Adaptive Partitioning of Persistent Time Series: A Pre-Registered Protocol with Synthetic-Pilot Validation and a Boundary-Quality Methodology Amendment

Working-paper protocol report on a pre-registered benchmark of Hurst-aware adaptive chunk-boundary partitioning against fixed-interval (TimescaleDB-style) and variance-aware (CUSUM) baselines. The pre-registration is archived separately on Zenodo (concept DOI 10.5281/zenodo.20188013). This preprint reports: (i) the registered protocol; (ii) a synthetic-pilot validation on a 16-series fractional Gaussian noise corpus that confirms the I/O accounting harness end-to-end and surfaces a methodological vulnerability in the primary outcome metric (bytes-read rewards aggressive subdivision on noise); (iii) Amendment 1 to the protocol, which adds a boundary-quality secondary outcome (F1 against ground-truth regime breaks on a new regime-switching synthetic corpus D4) without altering the primary outcome or the falsification rule. The empirical run on the registered datasets (S&P 500 50-ticker OHLCV and the Numenta Anomaly Benchmark) is gated on a v1.0 protocol lock in 2026-07 and will be reported separately. The artifact aims at the credibility-strengthening role increasingly played by pre-empirical protocol reports on arXiv. Cross-references to the author's other Zenodo artifacts: Malempati Hari (2026), Static and Temporal Fractal Coupling Between Volatility and Trading Volume (concept DOI 10.5281/zenodo.19611543); and Malempati Hari (2026), Multi-Scale Structural Descriptors for Governance-Relevant Patterns in Data Lineage Graphs (concept DOI 10.5281/zenodo.20099999).