Earlier pulse data gave uncertain prediction gains for 19 repeating CHIME/FRB sources

We tested whether earlier pulse measurements improve later measurement estimates in CHIME/FRB Catalog 2, release 20251231. We joined component rows into events and events into source-days. A fixed cohort had 19 repeating sources, 174 early training days, and 181 later test days. A common linear model used measured signal strength, dispersion measure, declination, time, and flagged-channel fraction. The source model added each source's mean early residual. Each source had equal weight in the final mean squared log error. The source model reduced estimated width error by 17.2% and spectral-extent error by 13.7%. Individual 97.5% source-bootstrap intervals were -14.2% to 42.1% and -36.7% to 46.5%. Thus neither primary test established lower error across sources. The intervals are approximate and conditional on fixed fitted predictions. Width error increased in the first-third/last-third check. Both point gains fell to about 3–5% after a 180-day gap. We report every specified sensitivity test. The package preserves code, provenance, the internal analysis plan, derived records, predictions, bootstrap samples, figures, and independent arithmetic checks. This is a retrospective test with measured burst covariates, not a forecast before detection. Catalog selection, propagation, the telescope beam, and spectral band limits prevent a claim about intrinsic emission physics.

fast radio bursts · CHIME/FRB · repeating sources · pulse width · spectral extent · chronological validation · source bootstrap · reproducible benchmark