ALP-ttbar Toy MC v2.7 — Post-Registration: Tag Formula Diagnostics Results in Displaced Light ALP → e⁺e⁻
Post-registration results for the v2.7 Tag Formula Diagnostics study in the ALP-ttbar Toy MC series. This study investigates why the Tag discriminator's upper-limit degradation ratio exhibits 45× higher seed-to-seed variability than the HV discriminator in a toy Monte Carlo simulation of displaced ALP → e⁺e⁻ searches in tt̄ events (E3 configuration: f = 0.25, 17-band endcap geometry, η = 2.0). Three diagnostic approaches were deployed across a 50-seed ensemble at two ALP masses (17 MeV and 200 MeV): Approach A — Frozen-Seed Variance Decomposition: Group F (both seeds vary), Group C (conversion seed fixed), and Group S (signal seed fixed) isolate the dominant variance source. Result: 99.84% of Tag variance comes from conversion-background randomness (σ_S/σ_F = 0.9992), not signal randomness (σ_C/σ_F = 0.0432) — the opposite of the preregistered prediction. Approach B — Per-Vertex Weight Analysis: 4-way overlap classification of near-band-edge and inter-band-gap vertices at multiple distance thresholds D ∈ {1, 2, 5, 10, 20, 50} mm. The near-edge correlation with Tag was weak (|R| = 0.0934, below the predicted 0.3 threshold), refuting prediction N1. Approach C — Robust Estimator Comparison: Trimmed mean at k = 1% reduces CV from 5.80% to 0.13% (a 45-fold improvement). Winsorised mean unexpectedly worsens CV to 7.35%, a counter-intuitive asymmetry caused by threshold-scan instability. Excess kurtosis = 44.97 confirms extremely heavy-tailed λ distribution. Prediction verdicts: 8/12 confirmed, 4/12 refuted. Core score: 2/4 (weak diagnostic evidence by preregistered criteria). The core refutations (F1 and N1) are scientifically informative — conversion dominance was unexpected and reshapes the understanding of Tag instability. At 200 MeV, variance additivity does not hold (σ_C²/σ_F² + σ_S²/σ_F² = 1.5975 ≠ 1.0); however, the departure from additivity alone cannot determine whether the signal and conversion sources are statistically dependent. The v2.7 Group F ensemble was computed fresh for this study. Seed-0 verification confirms internal consistency among the v2.7 F/C/S groups (matching to floating-point precision at i = 0), but direct comparison with the original v2.6 per-seed arrays was not performed (they were not retained). Preregistration DOI: 10.5281/zenodo.24388228Previous study (v2.6 post-registration): 10.5281/zenodo.24373115 This is an independent personal toy study, not affiliated with the authors of arXiv:2610.38932 (Alimena et al., 2026) or arXiv:2306.08686 (Rygaard et al., 2023), nor with the CMS, ATLAS, LHCb or ALICE Collaborations or the FCC study. Detector numbers are assumed; only qualitative trends and ratios are meaningful.
Authors
- Won Shik Paik
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-21
- DOI
- https://doi.org/10.5281/zenodo.24408989
- Primary Topic
- Particle Detector Development and Performance
- Type
- preprint