Distribution of magnitudes and estimated glint periods for our satellite dataset. The central scatter plot presents the full sample, with marginal histograms (blue: full sample, black outline: test subset). Colored lines show classification accuracy per model, ranging from 0 (no recovery) to 1 (perfect recovery). Due to the limited 63″ stamp FoV, glint periods ≳30″ are incompletely sampled (shaded region, see Sect. 4.2). The full model shows consistently high accuracy, whereas the cropped-16 model accuracy drops notably at ∼19.5 mag. FFT-based models show no glint-period bias, with only minor accuracy reductions at faint magnitudes.