当前设计：四块拼图The current design: four pieces

1. I001：一次切换一步1. I001: switch one step at a time

核心实验设计：在同一批 AO-26 星系上，把我们的处理方式逐步换成 eROSITA 叠加的方式，每次只换一步，记下每步把"SB/预测"移动了多少。六个候选步骤（按估计大小排序）：The core experiment: on the same AO-26 galaxies, change our processing step by step into the eROSITA-stack way, one step at a time, and record how much each step moves "SB/prediction". Six candidate steps (ordered by estimated size):

1. 源区：盘外"高"框 \(\to\) 包含盘的方位环；source region: off-disk "height" boxes \(\to\) azimuthal annulus including the disk;
2. 点源掩模：XMM 深度 \(\to\) eRASS1 等效深度；point-source mask: XMM depth \(\to\) eRASS1-equivalent depth;
3. PSF：把 XMM 图像与 eROSITA PSF 卷积、放在 36 Mpc 与 z = 0.08；PSF: convolve the XMM images with the eROSITA PSF, placed at 36 Mpc and z = 0.08;
4. 背景：局部环 \(\to\) 模型绝对背景；background: local annulus \(\to\) model absolute background;
5. kT：0.2 / 0.3 / 0.6 keV 之间的转换；kT: conversion between 0.2 / 0.3 / 0.6 keV;
6. 平均：反方差 \(\to\) 算术平均。averaging: inverse-variance \(\to\) arithmetic mean.

成功判据（02_CURRENT_QUESTION）：一张表，列出每步切换后 10–30 kpc 的测量值（零信号偏置已减），要么到达叠加水平，要么证明到不了。所有切换的结果目前待定（I001 只列出了清单，没有跑过）。Success criterion (02_CURRENT_QUESTION): a table of the measured 10–30 kpc value under each switched step (null-signal offset subtracted), that either reaches the stacked level or shows it cannot. All switch results are currently pending (I001 lists them; none has been run).

来源：cognition-os/insights/I001（方法对比表、候选原因、切换清单）；cognition-os/02_CURRENT_QUESTION.md（成功判据）。Source: cognition-os/insights/I001 (method table, candidate causes, switch list); cognition-os/02_CURRENT_QUESTION.md (success criterion).

2. I004：零信号对照叠加2. I004: the null-signal control stack

已有的 XMM 控制（27–50 kpc 盘上方空白框：底 \approx 1%，95% \le 3.5%）不在 CCD1 中心——而每个目标都坐在 CCD1 中心。I004 的设计：用目标在视场中心但没有延展发射的指向场（暗场 + 点源场），按同样流程处理，直接测出那个位置的仪器"晕"。六步：Existing XMM controls (blank boxes 27–50 kpc above the disk: floor \(\approx 1\)%, 95% \(\le 3.5\)%) are not on the CCD1 centre — where every target sits. I004's design: take pointings whose target is at the field centre but has no extended emission (dark + point-source fields), process them identically, and measure the instrument-made "halo" at that position directly. Six steps:

1. 池子：暗目标/空白场（仪器 + 天空）+ 点源目标（恒星、高红移 QSO：也含 PSF 翼与 OOT 条纹，匹配 AGN 主导的 ADAP 目标）；排除延展结构附近（MCXC、XCLASS、eRASS 延展源）。Pool: dark targets/blank fields (instrument + sky) + point-source targets (stars, high-z QSOs — also PSF wings and out-of-time streaks, matching the AGN-dominated ADAP targets); exclude fields near extended structure (MCXC, XCLASS, eRASS extended sources).
2. 按滤光片、模式、MOS1-CCD 损失历元、年份（QPB、SWCX）、干净曝光、SP 水平、银河前景 1:n 匹配。Match each science galaxy 1:n on filter, mode, MOS1-CCD-loss epoch, year (QPB, SWCX), clean exposure, SP level, Galactic foreground.
3. 把区域按星系距离从 kpc 复制到角分，把侧向框旋转到同一探测器角。Copy the regions from kpc to arcmin at the galaxy's distance; rotate edge-on boxes to the same detector angle.
4. 同一引擎（锁定的 esas_data_reduction）、掩模与 K 估计量。Same engine (the pinned esas_data_reduction), masks and K estimator.
5. 与科学权重组合；重采样配对。Combine with the science weights; resample pairings.
6. 按 MOS/pn、CCD1-only vs 跨 CCD、SP 水平、历元拆分，找原因。Split by MOS/pn, CCD1-only vs cross-CCD geometry, SP level, epoch to identify the cause.

试点的结果见"现状"页：pn \approx 0；MOS 偏置随场而变，原因开放。零场选择规则（历元、滤光片、曝光、NH、干净中心）与五个被淘汰的场见 null_pilot/selection/。The pilot's results are on the State page: pn \(\approx 0\); the MOS offset is field-dependent, cause open. The null selection rules (era, filter, exposure, NH, clean centre) and the five rejected fields are in null_pilot/selection/.

来源：cognition-os/insights/I004（设计六步）；XMMproposal2026 I020（1%/3.5% 底）；null_pilot/selection/SELECTION.md、SELECTION_v2.md。Source: cognition-os/insights/I004 (the six design steps); XMMproposal2026 I020 (1%/3.5% floor); null_pilot/selection/SELECTION.md, SELECTION_v2.md.

3. I007：普查式样本设计3. I007: the census-based sample design

样本设计不再凭经验，而是先"数一数档案里有什么"。覆盖规则：星系的整环要落在视场内（偏移 + 环角半径 \(\le\) 视场可用半径：ACIS-I 8.45′、ACIS-S 4.15′、XMM 14′）；指向取目标坐标（不是 aimpoint），忽略 roll（约 \pm20% 的不确定度）。Sample design no longer guesses; it first counts what is in the archive. Coverage rule: the galaxy's full circle must fit inside the field (offset + annulus angular radius \(\le\) the usable field radius: ACIS-I 8.45′, ACIS-S 4.15′, XMM 14′); pointings use target coordinates (not the aimpoint), roll ignored (about \pm20% uncertainty).

四条解读（普查者的结论，Provisional）：Chandra 适合 Zhang 式远距离样品；直接 Vrot 星表都在 z \lesssim 0.06，Vrot 选样的 Chandra 叠加在 150–250 kpc 只有几十颗；Vrot 假设用 eRASS1 公共数据检验；Chandra 的独特角色是点源深度。Four readings (the census-taker's conclusions, Provisional): Chandra suits the Zhang-like distant sample; direct-Vrot catalogues reach only z \(\lesssim 0.06\), so a Vrot-selected Chandra stack has only tens of galaxies at 150–250 kpc; the Vrot hypothesis is tested on eRASS1 public data; Chandra's distinctive role is point-source depth.

来源：cognition-os/insights/I007（(b) 覆盖规则、(c) M* 选样、(d) Vrot 选样、(e) 解读）。Source: cognition-os/insights/I007 ((b) coverage rule, (c) M*-selected, (d) Vrot-selected, (e) reading).

4. 研究计划 P1–P4、S1–S24. The research plan P1–P4, S1–S2

来源：cognition-os/RESEARCH_PLAN.md；cognition-os/SEARCH_QUEUE.md。Source: cognition-os/RESEARCH_PLAN.md; cognition-os/SEARCH_QUEUE.md.