NEW(W)曝光归一化残余的三条判别:值—向解耦、口径不沉性与累积指纹(Three Discriminants for Exposure-Normalised Residuals: Value–Direction Decoupling, Weight Robustness, and the Accumulation Fingerprint)

在扫描型探测器的曝光归一化分析中,残余计数率 r = obs/geo 通常被当作一个单一的幅度读数使用。本文指出,这种用法暗含三个未加检验的前提,并给出三条可独立陈述的判别,逐一检验它们:其一,值—向解耦(幅度与相位是否独立,相位漂移是否可归因于曝光几何自身的漂移);其二,口径不沉性(读数随模型选择改变、但不随分析手段改变的性质是否成立);其三,累积指纹(残余在分段时间序列上是往复还是累积)。以 DAMPE(暗物质粒子探测卫星)伽马射线事例样本(2016–2024,E ≥ 5 GeV,共 154,770 个事例)为主用例,以 Planck 2018 温度功率谱的峰位间距为对照用例,三条判别均给出明确判词:相位与幅度解耦(bootstrap corr(A, cos ψ) = +0.06),而相位漂移与几何曝光相位漂移高度相关(corr = 0.983),故相位不可读作被测物理量;共模幅度随天空口径落在 [3.79%, 9.94%],但在样本、时间与分辨率三类手段变换下不消亦不增(对半样本比值 0.997/1.004,模板平滑至 16° 仅下沉 5.7%);逐年幅度不随年数增长(8 个完整年 7.52% ± 1.29%,趋势 +0.10%/yr,χ²/dof = 1.11)。对照用例给出的是上界而非幅度,说明"测不出"与"不存在"在方法上必须分开。三条判别各自附有失效条件,可供同类分析直接取用或拒绝。 本文为方法论文:不重报观测发现,所引数据结果见已发表的发现报告(DOI: 10.5281/zenodo.23294277);不主张任何分量的物理身份;不主张三条判别与任何数学框架之间的推导关系(附录 A 只作结构对应的登记)。 关键词:曝光归一化;残余分析;值—向解耦;标架漂移;口径依赖;稳健性;累积指纹;DAMPE;CMB 峰位间距;方法学;初步分析。 — 老陈与AI的深夜实验室 发布 请笑纳 — In exposure-normalised analyses of scanning-type detectors, the residual count rate r = obs/geo is usually used as a single amplitude reading. It is pointed out here that this usage implicitly contains three untested premises, and three independently statable discriminants are given, by which each of them is tested in turn: first, value–direction decoupling (whether the amplitude and the phase are independent, and whether the phase drift can be attributed to the drift of the exposure geometry itself); second, weight robustness (the invariance of a residual under analysis operations), namely whether the property holds that the reading changes with the model selection but not with the analysis operations; third, the accumulation fingerprint (whether the residual, over a segmented time series, is reciprocating or accumulative). Taking the DAMPE (Dark Matter Particle Explorer) gamma-ray event sample (2016–2024, E ≥ 5 GeV, 154,770 events in total) as the primary use case and the peak spacing of the Planck 2018 temperature power spectrum as the control use case, all three discriminants yield explicit verdicts: the phase and the amplitude are decoupled (bootstrap corr(A, cos ψ) = +0.06), while the phase drift is highly correlated with the phase drift of the geometric exposure (corr = 0.983), so that the phase cannot be read as the measured physical quantity; the common-mode amplitude varies with the sky weight over [3.79%, 9.94%], yet it neither vanishes nor grows under transformations of the sample, of time, and of resolution (half-sample ratio 0.997 / 1.004; a 5.7% drop when the template is smoothed to 16°); and the year-by-year amplitude does not grow with the number of years (8 complete years, 7.52% ± 1.29%; trend +0.10%/yr; χ²/dof = 1.11). The control use case yields an upper bound rather than an amplitude, showing that "not detected" and "not present" must be kept apart methodologically. Each discriminant is accompanied by its failure conditions, so that an analysis of the same kind may adopt or reject it directly. This is a methodology paper: it does not re-report observational findings, which are given in the published discovery report (DOI: 10.5281/zenodo.23294277); it claims no physical identity for any component; and it claims no derivation between the three discriminants and any mathematical framework (Appendix A records structural correspondences only). Keywords: exposure normalisation; residual analysis; value–direction decoupling; frame drift; sky-weight dependence; robustness; accumulation fingerprint; DAMPE; CMB peak spacing; methodology; preliminary analysis. — Published by Lao Chen & AI's Late Night Lab. Please accept with a smile. —