A Pre-registered Test of a Stress Accumulation Model for Global Seismic Hazard Forecasting: Initial Validation of June–December 2026 Predictions

Earthquake forecasting remains one of the most challenging problems in Earth science, and progress has been hindered by the difficulty of distinguishing genuine predictive skill from hindsight bias. This study adopts a pre-registration framework: all model equations, parameters, and predictions were publicly timestamped on Zenodo on May 28, 2026, before the forecast window opened. The Encrypted Chain Fracture Model derives recurrence interval (T = S_crit / α) and magnitude (M = (2/3)·log₁₀(η·S_crit·L·W) − 6.0) from the deterministic relationship between tectonic stress accumulation and critical fracture along plate boundaries, using four parameters calibrated from publicly available GPS, seismic, and geological data. The model identifies eight critically stressed plate-boundary systems worldwide for June–December 2026. Following pre-registration, two of the predicted systems experienced large earthquakes within the forecast window: the Philippine Sea Plate convergent system (M7.9, June 8, 2026) and the New Britain arc system (M6.4, July 13, 2026). Both occurred within the predicted systems and time windows, with magnitudes broadly consistent with predictions. We evaluate predictive performance using the Molchan diagram, ROC analysis, binomial tests, probability gain, and gambling score against a null hypothesis of random guessing. The probability of obtaining two hits in two target earthquakes under random guessing with an alarmed fraction of 0.16 is P = 0.0256, significant at the α = 0.05 level. The probability gain per alarm is 6.25, indicating that the model provides substantial information beyond the background rate. Sensitivity analysis using an independent alternative target space classification confirms the robustness of this conclusion. Prospective Monte Carlo simulation of the remaining forecast window provides a quantitative framework for interpreting the final evaluation results. The model's physical assumptions are independently supported by GPS observations of linear strain accumulation, paleoseismic recurrence intervals, and historical earthquake catalogs. This is an interim evaluation based on approximately two months of the seven-month forecast window. A complete evaluation will be published at the conclusion of the forecast window on December 31, 2026, following a pre-specified format, irrespective of the outcome. The pre-registration approach, complete parameter calibration chain, transparent error analysis, pre-specified bias analysis protocol, and pre-committed final evaluation format provide a framework for falsifiable earthquake forecasting.