First-burst morphology predicts active FRB repeaters in CHIME Catalog 2, but eight published hidden-repeater lists did not predict which Catalog 1 one-off sources repeated

Question: do morphology-based predictions of hidden fast radio burst (FRB) repeaters come true in later data from the same telescope? Approach: We used CHIME/FRB Catalog 1 (to 2019-07-01) and Catalog 2 (to 2023-09-15). Part A: we extracted eight published machine-learning candidate lists made from Catalog 1 and compared later repetition of 467 Catalog 1 one-off sources. Part B: we trained classifiers on Catalog 1-era bursts with Catalog 1 labels only. Then we scored the first burst of 3,123 sources first detected after Catalog 1, with control for follow-up exposure, follow-up time, S/N and DM. Results: In Part B the score predicted a later burst: odds ratio 1.57 per SD (95% CI 1.40-1.76), AUC 0.76, stratified permutation p < 1e-4. The top score quintile held 34 of 54 repeating sources against 11.2 expected from exposure alone. The prediction applied to active repeaters: AUC 0.89-0.91 for sources with 3 or more events, 0.55-0.61 for sources with exactly 2 events. The score correlated with burst rate (Spearman 0.43-0.52). First bursts were as repeater-like as later bursts. In Part A, 12 of 467 one-off sources repeated, and 5 of them were not in the 2023 CHIME repeater paper. No list passed a Bonferroni correction (best one-sided Fisher p = 0.016 and 0.043). On the 5 strictly out-of-sample repeats every list agreed with no gain. Chen et al. 2022 listed 188 candidates; 8 of their listed one-off sources (4.5%) repeated against 1.4% of other sources. One Catalog 1 one-off on all eight lists (FRB 20190329A) is a burst of the known repeater FRB 20180814A. Limitations: few events in Part A (12 and 5), only 20 two-event repeaters in Part B, possible repeater-aware burst fits in Catalog 2, and no control for localization quality. Conclusion: morphology finds active repeaters, but these data do not support morphology-based claims that 30-60% of one-off sources are hidden repeaters.

fast radio bursts · CHIME/FRB · repeating FRB · hidden repeaters · machine learning · forecast test · burst morphology · Catalog 2 · out-of-sample validation