当前设计：v19 逐 ObsID 模型与质量筛选Current design: the v19 per-ObsID model and the quality mask

v19 Sherpa 模型（每个 ObsID 一次拟合）The v19 Sherpa model (one fit per ObsID)

对每个 ObsID，脚本把源谱（obj.pi，已减 QPB）与一个「天体 + 仪器」复合模型比对。天体部分（乘以天空面积 skyarea，单位 arcmin²）是：

For each ObsID the script compares the source spectrum (obj.pi, QPB-subtracted) with a compound "sky + instrument" model. The sky part (multiplied by the sky area skyarea, in arcmin²) is:

skyarea * ( LocalBubble + O7 + O8 + FeL
             + MWabs * ( CXB_peg + MWhalo + MWhothalo + Srcabs*SrcApec ) )
  + rsp_flat( skyarea * SoftProton )
  + RSP( skyarea * (mos instrumental lines) )

各分量含义与参数状态（依据 background_spectrum_fitting_2T_basedon_obsid_v19.py 第 812–1471 行）：

The components and their parameter states (from lines 812–1471 of background_spectrum_fitting_2T_basedon_obsid_v19.py):

来源：background_spectrum_fitting_2T_basedon_obsid_v19.py 第 812–1471 行（本页写作时读取：LocalBubble 833–842、MWhalo 844–853 与 1409–1413、CXB_peg 876–887 与 1429–1435、MWhothalo 855–863 与 1415–1419、SoftProton 1014–1039、仪器线 915–1012、MWabs 890–893 与 1402–1403、模型表达式 1178–1250）。

Source: background_spectrum_fitting_2T_basedon_obsid_v19.py lines 812–1471 (read when this page was written: LocalBubble 833–842, MWhalo 844–853 and 1409–1413, CXB_peg 876–887 and 1429–1435, MWhothalo 855–863 and 1415–1419, SoftProton 1014–1039, instrumental lines 915–1012, MWabs 890–893 and 1402–1403, model expression 1178–1250).

拟合流程与统计Fitting flow and statistic

• 统计量 chi2gehrels（对低计数箱误差做了 Gehrels 修正），优化器 levmar（日志与 E003 的 5 阶段流程一致）。Statistic chi2gehrels (Gehrels-corrected errors for low-count bins), optimiser levmar (the log and E003 agree on the 5-stage sequence).
• 能带：先 notice 0.4–8.0，再剔除 1.4–2.0 keV；分组在 0.4/1.4/2.0/8.0 keV 强制边界，剔除带内分组归零。Band: notice 0.4–8.0 first, then exclude 1.4–2.0 keV; grouping forces boundaries at 0.4/1.4/2.0/8.0 keV and zeroes groups inside the excluded band.
• XSPEC 物理配置：angr 丰度表、vern 截面、APECROOT=apec_v3.0.9。XSPEC physics: angr abundances, vern cross-sections, APECROOT=apec_v3.0.9.
• 误差：保存局部 kT–norm 协方差，δ 法传播到吸收流量；不是 profile-likelihood 区间。Errors: the saved local kT–norm covariance propagates to absorbed fluxes by the delta method; these are not profile-likelihood intervals.

来源：logs_obsid_fitting_v19_.../0800730201.log（Statistic = chi2gehrels，本页写作时读取）；cognition-os/experiments/E003「Method」第 2–6 条与「Verification」。

Source: logs_obsid_fitting_v19_.../0800730201.log (Statistic = chi2gehrels, read while writing this page); cognition-os/experiments/E003 "Method" items 2–6 and "Verification".

质量筛选：从 22 到 14The quality mask: from 22 to 14

明确的失败：0800731301 与 0800732201（dof≤0）、0800732601 与 0800732701（fail-closed，Sherpa succeeded=False）；Q 剔除 0800730401/1001/2101；0800730101 只有 MOS2，仅作灵敏度参考。

Explicit failures: 0800731301 and 0800732201 (dof≤0); 0800732601 and 0800732701 (fail-closed, Sherpa succeeded=False); Q-based exclusions 0800730401/1001/2101; 0800730101 is MOS2-only and sensitivity-only.

来源：cognition-os/experiments/E003「Technical results」与「Method」第 8 条；cognition-os/insights/I012「Measurement」。

Source: cognition-os/experiments/E003 "Technical results" and "Method" item 8; cognition-os/insights/I012 "Measurement".

为什么只用 MOS（不用 pn）Why MOS only (no pn)

初步 pn 分支有 22 个源谱，但只有 21 个同分支 QPB 匹配，17/22 的净曝光 < 10 ks。XMM-ESAS 指南指出：pn 的角区（corner）统计有限、有 out-of-time 事件、对软质子耀斑更敏感，此情形下 pn QPB 归一可能有偏（甚至过减）。这是产品就绪度决定，不是说 pn 仪器本身不可用，也不是光学漏光（I016）。

The preliminary pn branch has 22 source spectra but only 21 same-branch QPB matches, and 17/22 cleaned exposures are < 10 ks. XMM-ESAS guidance: pn corner statistics are limited, out-of-time events contaminate, and pn corner data are more sensitive to soft-proton flares, so the pn QPB normalisation can be biased (including oversubtraction) in this regime. This is a product-readiness decision — not a claim that pn is unusable, and not an optical light leak (I016).

来源：cognition-os/insights/I016（22/21/17 三个数与 ESAS「The Corner Problem」引用）。

Source: cognition-os/insights/I016 (the 22/21/17 numbers and the ESAS "The Corner Problem" citation).

计划中的 E013 联合空间似然The planned E013 joint spatial likelihood

MW 与 M31 热发射共享 kT、丰度与吸收，谱形相同——只有空间形状能区分。因此：

MW and M31 thermal emission share kT, abundance and absorption, so they are spectrally identical — only the spatial shape separates them. Therefore:

• 阶段 A1（运行中）：对每个场，在 (kT, norm) 网格上冻结 CGMsum 分量、重拟合其余一切，得到 Δstat 曲面。Stage A1 (running): for each field, freeze the CGMsum component on a (kT, norm) grid and re-fit everything else, giving Δstat surfaces.
• 阶段 A2（Claude）：联合似然 Σ_fields Δstat_i(kT, norm_i = N_MW(b_i) + A·g(r_i; r_c, β))，对银河系项 N_MW（常数或线性 b）、M31 幅度 A 做 profile；A=0 边界用模拟校准单侧上限。Stage A2 (Claude): the joint likelihood Σ_fields Δstat_i(kT, norm_i = N_MW(b_i) + A·g(r_i; r_c, β)), profiling over the MW term N_MW (constant or linear in b) and the M31 amplitude A; one-sided limits on A calibrated at the A=0 boundary by simulation.

来源：cognition-os/experiments/E013_joint_spatial_profile_likelihood.md（设计全文）；.agent/jobs/2026-10-14-e013-joint-profile/BRIEF.md（A1 网格与验收）。

Source: cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (the design); .agent/jobs/2026-10-14-e013-joint-profile/BRIEF.md (the A1 grid and acceptance).

6. 动画：谱的叠加6. Animation: stacking the spectrum

这是一个玩具示意（非真实响应）：勾选各分量、拖动 QPB 残差，看总谱如何形成。真实的逐 ObsID 拟合见上文模型表。

This is a toy illustration (not the real response): tick the components and drag the QPB residual to see how the total forms. The real per-ObsID fit uses the model table above.

来源：曲线为教学示意（本页自带 JS，无外部数据）；分量集合与剔除带对应 v19 模型（background_spectrum_fitting_2T_basedon_obsid_v19.py）。

Source: the curves are a teaching toy (this page's own JS, no external data); the component set and the excluded band follow the v19 model (background_spectrum_fitting_2T_basedon_obsid_v19.py).