Reproducibility package: Flow-like onset events co-occur with a stability threshold across noise conditions: single-observer study without evidence of critical slowing down

This repository contains all raw data and analysis code required to reproduce the quantitative findings, figures, and tables reported in the manuscript: "Flow-like onset events co-occur with a stability threshold across noise conditions: single-observer study without evidence of critical slowing down" Yosuke Yoshimura (2026) Contents: - data/: Raw observation logs from experiments exp13–exp17 (JSON format, 17 files). Each file is an array of flow-onset event objects containing clusterStability, directionCoherence, beatPhase, noise, fade, pressMode, pre300ms_delta, csd_buf_length, csd_var_slope, and csd_ac1_slope. - scripts/: Four analysis scripts (Python 3): empirical_figures.py (Figures 1–2, claims C1–C3, C6), exp17_analysis.py (Figure 3, claims C8–C13), csd_analysis.py (Figure 4, claims C4–C5), stability_model.py (Figures 5–6, Supplementary Figure S1, claim C7). - figures/: Submission-quality PNG figures (Figures 1–6, Supplementary Figure S1). - source_data/: Per-figure source CSV files for all 7 figures (Figures 1–6, S1). - README_FOR_REPOSITORY.md: Quick-start guide, data schema, session selection rules, claim audit summary. All 13 quantitative claims in the manuscript have been verified against raw data. Pearson r = −0.978 for the dirCoh–noise correlation (Figure 2, claim C2) is computed on 6 per-bin mean values, not on 209 individual events; this is documented in the README and stated explicitly in §2.8 of the manuscript. Software: Python ≥ 3.8, numpy ≥ 1.21, scipy ≥ 1.7, matplotlib ≥ 3.4. No proprietary software required.