Threshold-Based Inference Under Finite Access: Four Controlled Experiments on Scale-Dependent Decision Instability

Four controlled experiments comparing a standard threshold-based inference rule against a scale-invariance check on real ECG time-series data (MIMIC-Perform). All experiments use identical, locked parameters with no tuning or case-specific adjustment. Results identify three structurally distinct failure modes of single-valuation inference under finite access, including cases where a raw metric overrides stable multi-scale structure and cases where legitimate smoothing scales never converge to a single outcome. Includes full audit artifacts (CSVs, plots, parameter files, null permutation tests) for independent inspection. See EXECUTIVE_SUMMARY.txt for an overview and EXPERIMENTAL_REPORT.txt for full details.