Code and data for: Leakage-audited machine learning versus ETAS for earthquake forecasting in the Sea of Marmara
Basri Kerem Alhan, Kenessary Khabat
Code, processed data products, configuration, and results artifacts for "Machine learning versus ETAS for earthquake forecasting in the Sea of Marmara: a leakage-audited negative result and a closed-form scoring artifact" (Alhan & Khabat, submitted to Seismica). Version 1.2.0 accompanies the Seismica resubmission: the registered count-scored evaluation is re-adjudicated under a proper binary-occurrence score with a closed-form identity for the scoring artifact a(h) = h − 1 − ln h; the operational b_op = 1.15 is restated as a convention with full forensics; a feature-ablation and grouped-PCA study locates all ranking information on the ETAS axis; and the repository is organized by pipeline stage.The archive carries the machine-readable claims files (claims.json, the registered adjudicator of record; round3/claims_bernoulli.json; round4/claims_sensitivities.json), the block-bootstrap intervals, the dated, hashed pre-registration and amendment chain (docs/preregistration/, with its hash audit in results/audit/preregistration_chain.json), the placebo-battery outputs, the pyCSEP inputs and results, and the reproduce-all target (scripts/release/reproduce_all.py), whose 23 artifact assertions pass in this distribution as shipped.The processed catalogue is a derived dataset redistributed with attribution to Boğaziçi University KOERI-RETMC (see DATA_LICENSE.md). GNSS velocities: Nevada Geodetic Laboratory. Fault model: GEM Global Active Faults Database. Code is MIT-licensed. Development repository: https://github.com/keremalhan/marmara-forecast