Data and code for: An empty calibration window: why magnitude-threshold traffic-light schemes cannot warn of moderate induced earthquakes in the southern Sichuan Basin

Data and analysis code accompanying the manuscript “An empty calibration window: why magnitude-threshold traffic-light schemes cannot warn of moderate induced earthquakes in the southern Sichuan Basin” (Qi Deng, submitted to Journal of Seismology). The study asks how much advance warning a conventional magnitude-threshold traffic-light scheme (TLS) could have delivered before the moderate earthquakes of the southern Sichuan Basin between 2015 and 2024, and tests seven published and bounding designs against the full 2010–2024 open record. The deposit contains the earthquake catalogues used, the derived event lists, the development-zone geometry, and the complete Python analysis code, including the traffic-light back-test, the goodness-of-fit and sensitivity analyses, and the figure generator. Contents Mapped fault traces shown in Fig. 1 are extracted from the China Active Faults Database (1:4 000 000; Wu et al. 2024, Earth Syst. Sci. Data 16, 3391–3417, doi:10.5194/essd-16-3391-2024) and shipped in ESM_4.zip. ESM_1.csv (Online Resource 1) – USGS PDE catalogue for the Yibin region, 2010-01-01 to 2024-12-31, M ≥ 3.0 (239 events after removal of non-earthquake records) ESM_2.txt (Online Resource 2) – ISC Bulletin agreement-check counts by year, 2010–2024 ESM_3.csv (Online Resource 3) – complete M ≥ 4.5 event list (117 events) ESM_4.zip (Online Resource 4) – catalogue download, completeness, phase-rate and sensitivity scripts, the mainshock-migration test, and the figure / graphical-abstract generators ESM_5.json (Online Resource 5) – approximate development-zone centres and bounds used for all distance statistics ESM_6.zip (Online Resource 6) – the complete traffic-light back-test (Sections 4.4, 4.5–4.9 of the manuscript), with its own README containing full definitions and run instructions SUPPLEMENTARY_README.txt – file manifest Traffic-light back-test (ESM_6). S6_tls_backtest.py defines the seven designs, builds the daily traffic-light state series, and performs the back-test and the Gutenberg–Richter extrapolation; S7_tls_figure.py reproduces Fig. 5; S8_spatial_buffer.py carries out the spatial-buffer sensitivity of Section 4.7 — the scheme armed only on events within a radial buffer around each development zone, at R = 5, 10, 15, 20 and 30 km; S9_sensitivity.py runs the magnitude-scale-offset sweep and the 2019-01-03 Gongxian sensitivity; and S10_bvalue_gof.py performs the goodness-of-fit test of the fitted Gutenberg–Richter law (Kolmogorov–Smirnov with a Monte-Carlo, parameter-estimation-corrected p-value whose null distribution is simulated at the catalogue's 0.1 magnitude reporting precision) together with the declustered b-values of Section 4.9. Every script reads only from data/ and writes only to out/; the machine-readable results underlying Fig. 5 and Tables 6–7 are included. Runtime is a few seconds per script and no network access is required. Key results reproduced here. Every published design whose red threshold lies at or below ML 4.0 collapses to a single rule on an M ≥ 4.0 catalogue (red on 62.8% of armed days, 38 stop-work episodes in ten years). Raising the threshold to the highest still enforceable value (M 5.0) removes the saturation but leaves four of the seven armed-period mainshocks without any live escalation. Restricting the scheme to the 10–20 km footprint at which these protocols are enforced lowers saturation to 11.5–20.7% but leaves site-specific capture at 43%, so the failure mode moves from false alarms to missed events rather than disappearing. No red threshold between M 4.0 and M 6.0 achieves both 50% precursor capture and a 30-day exceedance interval. Data sources. The raw catalogues are openly available from the USGS PDE (earthquake.usgs.gov/fdsnws) and the ISC Bulletin (fdsnws.isc.ac.uk); they are redistributed here for reproducibility. The China Earthquake Networks Center catalogue was not accessible under the author's institutional data-access constraints; where Chinese-network magnitudes are quoted they are taken from published values and are listed in Table 3 of the manuscript. Software. Python 3.13 with NumPy, pandas, SciPy and matplotlib. License: CC BY 4.0.