ALP-ttbar Toy MC v1.4: Post-registration Results for Bremsstrahlung-Aware Vertex Fits (Inner-Hit Fit and Gaussian Brems-Noise Fit) of Displaced Light ALP → e+e−

Motivated by the displaced-dielectron extension (arXiv:2610.38932, Alimena et al.) of the top-associated long-lived ALP strategy (arXiv:2306.08686, Rygaard et al.), this independent toy series studies detector-level effects that the generator-level projections of arXiv:2610.38932 do not include. Not affiliated with the authors or the CMS Collaboration; only qualitative trends are meaningful, and no claim is made about the performance of a real detector. v1.4 tests two brems-aware vertex fits on the identical simulated hits used in v1.3, under a public preregistration: (a) an inner-hit fit and (b) a Gaussian brems-noise fit (single-Gaussian approximation of a Gaussian-sum filter). The reference points are the v1.3 brems-blind fit (178x loss in reach at 17 MeV) and the no-brems bound (about 2x). Fit (b) repairs the calibration (bias +1 mm, unit pulls) but not the resolution (core about 7-11 mm instead of about 2 mm), so the loss improves only to 126x at 17 MeV; fit (a) gives 169x. Three of five preregistered predictions were refuted, two confirmed. At 100-200 MeV the remaining loss is dominated by the conversion constraint being applied to ALP pairs. Next steps: a Gaussian-sum filter and a signal-appropriate vertex hypothesis. Code: github.com/wspaik5/ALP_ttbar_Toy_MC_v1.4_Postregistration_Brems_Aware_FitPreregistration: github.com/wspaik5/ALP_ttbar_Toy_MC_v1.4_Preregistration_Brems_Aware_FitPrevious versions: v1.3 10.5281/zenodo.23862775, v1.2 10.5281/zenodo.23847880, v1.1 10.5281/zenodo.23833475, v1.0 10.5281/zenodo.23825276