<p>Magnetocardiography (MCG) provides non-contact, radiation-free recordings of cardiac magnetic activity, but its clinical translation has been hindered by the lack of standardized and interpretable visual criteria. Here we propose an interpretable visual computing framework that converts one-dimensional MCG timeline maps (TLMs) into a standardized visual lexicon for coronary artery stenosis diagnosis. Inspired by the visual logic of electrocardiographic ischemia interpretation, we defined nine quantifiable TLM indices capturing amplitude, direction, and waveform-ratio abnormalities across a 36-channel MCG array. We evaluated this framework in 4,438 individuals, including 2,673 patients with angiographically confirmed obstructive coronary artery disease and 1,765 ECG-negative reference participants. Data from one hospital campus were used for model development and internal validation, while two independent cohorts were used for external validation. Vessel-specific random forest models integrating TLM indices and spatial channel information achieved robust diagnostic performance for LAD, LCX, RCA, and LM stenosis, with particularly strong performance in ECG-negative individuals. Visual–anatomical interpretation and SHAP analyses showed that the model relied on spatially localized waveform abnormalities consistent with coronary perfusion territories. These findings establish a reproducible visual lexicon for MCG-TLM interpretation and demonstrate its potential as an interpretable adjunctive tool for non-invasive coronary risk stratification.</p>