CGMStacking 工作台CGMStacking workbench
项目已暂停（D021，2026-10-22 23:42 EDT）。PI 确认目前没有可用的 eRASS:2 光子数据，所有执行任务已停止。本手册是收尾交接：读完"现状"与"如何接手"两页即可在将来无 AI 地恢复。恢复条件：eRASS:\(\ge 2\) 事件可用（DR3 约 2028 下半年），或 PI 另行决定。The project is paused (D021, 2026-10-22 23:42 EDT). The PI confirmed that no eRASS:2 photon data are available to us; every executor is stopped. This handbook is the closing hand-off: reading the "State" and "By hand" pages is enough to resume later without AI. Resume condition: eRASS:\(\ge 2\) events become available (DR3 around the second half of 2028), or the PI decides otherwise.
eROSITA 叠加——Zhang et al. 2024（SDSS 中央星系、银河系质量档）与 He & Li 2026（474 颗 L* 星系，15–50 Mpc）——在银河系质量星系周围 10–50 kpc 处测到热星系周介质（circumgalactic medium, CGM）的 X 射线辐射；而我们对手头邻近侧向旋涡星系的 XMM-Newton 逐星分辨测量（分档 4 星系、Vc 标度）只给出 SB/prediction \(\approx 0.17 \pm 0.10\)，低了约 3–6 倍。CGMStacking 的主线按 PI 决定演进了三次：先问是哪一步数据处理差别造成了这个矛盾（已停泊，XMM 线移交独立项目，D015/D017）；再问档案 Chandra 叠加能否测到热 CGM（2026-10-22 裁决：不能；改为 95% 上限，D020）；最后转向复现 eROSITA 叠加并检验 Chandra 在哪里改善它——可行性已完成（eRASS1 He & Li 复现 GO；Zhang eRASS:4 复现 NO-GO），随后因没有 eRASS:2 光子数据而暂停（D021）。旁支：旋转速度（Vrot）是否比恒星质量（M*）更好地预测热 CGM——eRASS1 检验已给出答案（不确定，负号）。eROSITA stacks — Zhang et al. 2024 (SDSS central galaxies, Milky-Way-mass bin) and He & Li 2026 (474 L* galaxies at 15–50 Mpc) — report hot circumgalactic-medium (CGM) X-ray emission at 10–50 kpc around Milky-Way-mass galaxies, while our resolved, galaxy-by-galaxy XMM-Newton measurements of nearby edge-on spirals (4 galaxies in the combination, Vc scaling) give SB/prediction \(\approx 0.17 \pm 0.10\), a factor \(\approx 3\)–6 lower. The project's main line evolved three times by PI decision: first which processing step accounts for this tension (now parked; the XMM line moves to a separate project, D015/D017); then can an archival Chandra stack measure the hot CGM (verdict on 2026-10-22: no; redirect to a 95% upper limit, D020); finally reproduce the eROSITA stack and test where Chandra improves it — feasibility done (the eRASS1 He & Li reproduction is a GO; the exact Zhang eRASS:4 reproduction is a NO-GO), then paused because no eRASS:2 photon data exist for us (D021). Side question: does rotation velocity (Vrot) predict the hot CGM better than stellar mass (M*) — the eRASS1 test has answered (ambiguous, negative sign).
来源：cognition-os/00_AI_BRIEFING.md；cognition-os/02_CURRENT_QUESTION.md；cognition-os/04_DECISION_LOG.md D015、D017、D020、D021。Source: cognition-os/00_AI_BRIEFING.md; cognition-os/02_CURRENT_QUESTION.md; cognition-os/04_DECISION_LOG.md D015, D017, D020, D021.
组会（最新一天）Group meeting (the latest day)
2026-10-23（UTC）——最后一天：上极限转向与暂停2026-10-23 (UTC) — the last day: the upper-limit turn and the pause
成果 / ResultsResults
- P65/P66（Chandra 背景残余）：空位差分残余随场变化——P53 场 blank-sky \(-18.6\%\) [\(-34.5, -7.5\)]，P58 目标场 \(+8.6\%\)/\(+12.3\%\)——量级 10–20%，不是单一常数（R095、R102，Provisional）。P65/P66 (the Chandra background residual): the empty-position differential residual is field-dependent — P53 fields blank-sky \(-18.6\%\) [\(-34.5, -7.5\)], P58 target fields \(+8.6\%\)/\(+12.3\%\) — at the 10–20% level, not a single constant (R095, R102, Provisional).
- P68：把 Chandra 预报按 heli474 10–30 kpc 锚重新归一后，最佳单元 S/N 1.17（无系统误差）\(\to\) 0.07（P65 包络）——档案 Chandra 叠加测不到 eRASS1 水平的 10–30 kpc CGM（R103，Provisional）。P68: re-anchored to the heli474 10–30 kpc anchor, the best-cell S/N is 1.17 (no systematics) \(\to\) 0.07 (P65 envelope) — an archival Chandra stack cannot detect the 10–30 kpc CGM at the eRASS1 level (R103, Provisional).
- P69（可行性）：eRASS1 He & Li 474 复现 GO（本地泊松 7.16\(\sigma\) / 按星系自助 3.98\(\sigma\)，论文 7.4\(\sigma\)）；Zhang eRASS:4 精确复现 NO-GO（DR2 只有目录；DR3 约 2028 下半年）。对 P58 试点的独立复核全点 HOLDS（R105、R097）。P69 (feasibility): the eRASS1 He & Li 474 reproduction is a GO (local Poisson 7.16\(\sigma\) / galaxy-bootstrap 3.98\(\sigma\), the paper 7.4\(\sigma\)); the exact Zhang eRASS:4 reproduction is a NO-GO (DR2 is catalogue-only; DR3 around the second half of 2028). An independent review HOLDS on all points of the P58 pilot (R105, R097).
工作 / WorkWork
- P58 半 1 到 35/213 个已验证 ObsID；P60 预注册接受（515 对、四个零门）。PI 决定 D020（上极限 + eROSITA 复现主线）、D021（暂停；P67/P70/P71/RT-31/RT-33 全停；P58 Virgo 驱动停止）。P58 half 1 reached 35/213 verified ObsIDs; the P60 pre-registration accepted (515 pairs, four null gates). PI decisions D020 (upper limit + the eROSITA-reproduction main line) and D021 (pause; P67/P70/P71/RT-31/RT-33 all stopped; the P58 Virgo driver stopped).
问题 / ProblemsProblems
- P58 进度曾被误报（manifest 计数 vs 真值 22 个 ObsID，R101）；驱动在 ObsID 22445 下载截断处崩溃（retry/skip 重启）；盲法两次被内部摘录破坏（已移入 blinded/）；对注入检验的红队 找到 P57 两处代码缺陷（曝光盲 \(\alpha\)、DETNAM 正则），修复任务被暂停终止——未修（R098）。P58 progress was mis-reported (manifest count vs a true 22 ObsIDs, R101); the driver crashed on a truncated download at ObsID 22445 (restarted with retry/skip); blinding was broken twice by an internal excerpt (moved to blinded/); the red team on the injection test found two P57 code defects (exposure-blind \(\alpha\), DETNAM regex), and the fix job was stopped by the pause — not fixed (R098).
待 PI 决定 / For the PIFor the PI
- 恢复时机：eRASS:\(\ge 2\) 事件可用（DR3 约 2028 下半年）或 PI 另行决定；恢复点见"如何接手"页。Resume timing: when eRASS:\(\ge 2\) events become available (DR3 around the second half of 2028), or the PI decides; resume points are on the "By hand" page.
完整两天记录（2026-10-22 与 2026-10-23 UTC）见 组会页。来源：cognition-os/04_DECISION_LOG.md D020–D021；07_REVISION_LOG.md R097–R105；06_AGENT_LOG.md。The full two-day record (2026-10-22 and 2026-10-23 UTC) is on the meetings page. Source: cognition-os/04_DECISION_LOG.md D020–D021; 07_REVISION_LOG.md R097–R105; 06_AGENT_LOG.md.
目标Objectives
| 问题Question | 状态Status | 当前最好数字Best current number | 趋向Trend |
|---|---|---|---|
| 原主问题（已停泊）：eROSITA 式叠加与我们分辨测量之间的因子由哪一步处理差别造成？Original main question (parked): which processing difference accounts for the factor between an eROSITA-style stack and our resolved measurements? | 停泊（暂停）Parked (paused) | 分辨组合 \(0.17 \pm 0.10\)（4 星系，Vc 标度；系统误差未传播）；工作区间 \(0.25\)–\(0.37 \pm 0.10\)（MOS 零值偏置等三项上修后）；几何解释已被排除（同 CGM 预期比 \(\approx 1.00\)，张力 \(0.18\)–\(0.48\)）。Resolved combination \(0.17 \pm 0.10\) (4 galaxies, Vc scaling; systematics not propagated); working range \(0.25\)–\(0.37 \pm 0.10\) after three upward corrections; the geometry explanation is ruled out (same-CGM expected ratio \(\approx 1.00\), tension \(0.18\)–\(0.48\)). 来源：03_TRUST_TABLE.md T02；R009、R010、R032、R033Source: 03_TRUST_TABLE.md T02; R009, R010, R032, R033 |
持平（停泊）No significant difference (parked) |
| 主目标（D020 演进）：用 P58 半 1 给出档案 Chandra 叠加的 10–30 kpc CGM 95% 上限；并复现 eROSITA 叠加、检验 Chandra 在哪里改善它。Main goal (as evolved by D020): report the 95% upper limit on the 10–30 kpc CGM from an archival Chandra stack using P58 half 1; and reproduce the eROSITA stack, testing where Chandra improves it. | 暂停（可行性已完成）Paused (feasibility done) | 上限机器未建成（P70 停止）；P58 半 1 已还原 35/213 个 ObsID，目标计数保持盲。可行性结论：eRASS1 He & Li 474 复现 GO；Zhang eRASS:4 精确复现 NO-GO（DR2 目录-only，DR3 约 2028 H2）；Chandra/CSC 覆盖 heli474 \(\ge 80\%\) 只有 30/474 \(\to\) 30 颗试点（P71，停止）。The upper-limit machinery is not built (P70 stopped); P58 half 1 has 35/213 ObsIDs reduced, the target counts still blinded. Feasibility: the eRASS1 He & Li 474 reproduction is a GO; the exact Zhang eRASS:4 reproduction is a NO-GO (DR2 catalogue-only, DR3 ~2028 H2); Chandra/CSC coverage of heli474 at \(\ge 80\%\) is only 30/474 \(\to\) a 30-galaxy pilot (P71, stopped). 来源：04_DECISION_LOG.md D020、D021；R103、R105；P69 REPORT.luna.mdSource: 04_DECISION_LOG.md D020, D021; R103, R105; the P69 REPORT.luna.md |
推进中（可行性 GO）Advancing (feasibility GO) |
| 子问题 A（已关闭，负结论）：档案 Chandra 叠加能否测到热 CGM？Sub-question A (closed, negative): can an archival Chandra stack detect the hot CGM? | 关闭（不可达）Closed (not reachable) | 关闭证据：heli474 锚重归一后最佳单元 S/N 1.17（DD0）\(\to\) 0.07（P65 包络）；此前 P44 全样本 S/N 2.22/1.49/0.90（三种噪声假设），无条件上限 5.16；MW 档 50–150 kpc 即使只算光子也只有 2.32/0.97。Closing evidence: re-anchored to the heli474 anchor the best-cell S/N is 1.17 (DD0) \(\to\) 0.07 (P65 envelope); before that P44 gave full-sample S/N 2.22/1.49/0.90 (three noise assumptions), conditional upper bound 5.16; the MW bin at 50–150 kpc is 2.32/0.97 even photon-only. 来源：R103、R069（P44）、R015（P13）；D020Source: R103, R069 (P44), R015 (P13); D020 |
关闭（负结论）Closed (negative) |
| 子问题 B（已回答）：旋转速度（Vrot）是否比恒星质量（M*）更好地预测热 CGM？Sub-question B (answered): does rotation velocity (Vrot) predict the hot CGM better than stellar mass (M*)? | 已回答：不确定（负号）Answered: ambiguous (negative) | eRASS1 预注册检验（D012 样本 4096 颗，10–50 kpc，等权平均 SB 估计量，\(b = 0.88\)）：\(\Delta V = -0.307\)，按星系自助 \(\sigma = 0.178\)，90% CI [\(-0.622, -0.088\)]（单侧 95% 上限 \(-0.088\)）；按 D013 分层判"不确定"（\(\Delta V < 0\)）。红队复核 HOLDS（不是少数星系伪影；去掉环内星表源后更负，\(-1.00\)，515 颗）。The pre-registered eRASS1 test (D012 sample of 4096 galaxies, 10–50 kpc, equal-weight mean SB estimator, \(b = 0.88\)): \(\Delta V = -0.307\), galaxy-bootstrap \(\sigma = 0.178\), 90% CI [\(-0.622, -0.088\)] (one-sided 95% upper limit \(-0.088\)); verdict "ambiguous" under the D013 tiers (\(\Delta V < 0\)). The red-team review HOLDS (not a few-galaxy artefact; removing galaxies with catalogue sources in the annulus makes it more negative, \(-1.00\), 515 galaxies). 来源：R060（P29）；R067（RT-21）；04_DECISION_LOG.md D013Source: R060 (P29); R067 (RT-21); 04_DECISION_LOG.md D013 |
已回答（不确定）Answered (ambiguous) |
趋向列对比上一次刷新（数据截至 2026-10-21 11:31 EDT）：原主问题持平；子问题 A 从"推进中"变为"关闭（负结论）"；子问题 B 从"待定"变为"已回答（不确定）"；新主目标（上极限 + eROSITA 复现）为新增行。The trend column compares with the previous refresh (data as of 2026-10-21 11:31 EDT): the original main question is flat; sub-question A moved from "advancing" to "closed (negative)"; sub-question B from "pending" to "answered (ambiguous)"; the new main goal (upper limit + eROSITA reproduction) is a new row.
未解决的问题（暂停时）Open problems (at the pause)
| 问题Problem | 为何重要Why it matters | 当前最佳假设 / 关闭证据Best hypothesis / closing evidence | 下一个检验（恢复点）Next test (resume point) | 状态Status |
|---|---|---|---|---|
| 1. eRASS1 He & Li 474 复现（新主线第一步）1. The eRASS1 He & Li 474 reproduction (first step of the new main line) | 它是"复现 eROSITA 叠加"这条主线的地基；也校准我们自己的叠加流水线。It is the foundation of the "reproduce the eROSITA stack" main line; it also calibrates our own stacking pipeline. | 本地已到计数层面：10–50 kpc 净计数 \(1751.5 \pm 244.6\)，泊松 7.16\(\sigma\)（论文 7.4\(\sigma\)）；但按星系自助只有 3.98\(\sigma\)，213/474 单星系净值为负——两种误差定义必须并列报告。Locally already at count level: 10–50 kpc net counts \(1751.5 \pm 244.6\), Poisson 7.16\(\sigma\) (the paper 7.4\(\sigma\)); but the galaxy bootstrap gives only 3.98\(\sigma\), with 213/474 single-galaxy net values negative — the two error definitions must be reported side by side. | 按 P69 顺序 1 重做并加 2.3–5 keV 控制带、随机场与背景区替换检查；先读 P69 REPORT。Redo following P69's order 1, adding the 2.3–5 keV control band, random fields and background-region swaps; read the P69 REPORT first. | 可行（GO），暂停Feasible (GO), paused |
| 2. Chandra 10–30 kpc 的 95% 上限（P58 半 1）2. The 95% upper limit on the 10–30 kpc CGM from P58 half 1 | D020 定下的新主目标；上限携带实测背景残余包络作为系统误差。The new main goal set by D020; the limit carries the measured background-residual envelope as a systematic. | P58 已还原 35/213 个 ObsID（0.213 GB/ObsID）；P60 预注册冻结（515 对）；第四个零门（共同残余）未过——目标计数保持盲。上限机器（P70）未建成即停。P58 has 35/213 ObsIDs reduced (0.213 GB/ObsID); the P60 pre-registration is frozen (515 pairs); the fourth null gate (the common residual) did not pass — the target counts stay blinded. The upper-limit machinery (P70) was stopped before it was built. | 重启 P58 驱动（sol/P58/scripts 的 retry/skip 驱动）补完半 1；再建 P70 机器；先读 P60 PREREG.md 与 R094。Restart the P58 driver (the retry/skip driver in sol/P58/scripts) to finish half 1; then build the P70 machinery; read the P60 PREREG.md and R094 first. | 暂停Paused |
| 3. Chandra 背景的内对外残余（限制性校准问题）3. The Chandra inner-versus-outer background residual (the limiting calibration problem) | 它把 P55 预报从 S/N 4.00 压到 0.96（blank-sky 包络）；不解开它，任何 Chandra 叠加的差分都带 10–20% 的场依赖误差。It squeezes the P55 forecast from S/N 4.00 to 0.96 (blank-sky envelope); unsolved, every Chandra-stack difference carries a 10–20% field-dependent error. | P65：P53 场 blank-sky \(-18.6\%\) [\(-34.5, -7.5\)]，stowed \(-17.4\%\)，场间 sd \(\approx 59\%\)，渐晕/双成分修正无效，常数不可迁移；P66：P58 目标场 \(+8.6\%\)/\(+12.3\%\)（分母不同）。读法：场依赖，不是单一偏移。P65: P53 fields blank-sky \(-18.6\%\) [\(-34.5, -7.5\)], stowed \(-17.4\%\), field-to-field sd \(\approx 59\%\), vignetting/two-component corrections do not help, constants do not transfer; P66: P58 target fields \(+8.6\%\)/\(+12.3\%\) (different denominators). Reading: field-dependent, not a single offset. | 对 P65 的独立复核红队（攻击 P65）被暂停终止——恢复后先补这个独立复核，再谈修正。the independent-review red team on P65 (attacking P65) was stopped by the pause — after resuming, run that independent review before any correction. | 进行中（暂停）Provisional (paused) |
| 4. eRASS1 负 \(\Delta V\) 的机制4. The mechanism of the negative eRASS1 \(\Delta V\) | 快转组的叠加面亮度反而更低；若非伪影，它约束"Vrot 预测 CGM"的物理。The fast-rotators' stacked surface brightness is lower; if not an artefact, it constrains the physics of "Vrot predicts the CGM". | 红队复核 HOLDS（背景平衡；去环内源后更负）。P42：快组星表 SFR 低 0.084 dex（SMD \(-0.22\)）；HI 质量与颜色在冻结表里没有——机制仍是假设。The red-team review HOLDS (backgrounds balanced; more negative after removing in-annulus sources). P42: the fast group's catalogue SFR is lower by 0.084 dex (SMD \(-0.22\)); HI mass and colour are absent from the frozen tables — the mechanisms remain hypotheses. | P42 建议的后续：能量分瓣 + HI 受控的复核（恢复后）。P42's proposed follow-up: an energy-resolved, HI-controlled recheck (after resuming). | 开放（暂停）Open (paused) |
| 5. 叠加代码缺陷（P57/P63）5. The stacking-code defects (P57/P63) | 上限与检出统计都跑在这套代码上；两处已知缺陷会偏置注入检验与门判定。Both the upper limit and the detection statistic run on this code; two known defects bias the injection test and the gate decision. | 对注入检验的红队：固定 \(\alpha\) 忽略曝光（零点 \(+23.3\) 计数，1.21\(\sigma\)）；DETNAM 正则使 9/102 ObsID 注入高估 \(\times 2.3\)。对门代码的审计红队：\(\sigma_{\rm total}\) 缺 V_common；门判定未在目标读取路径强制执行。修复任务 P67/P69 被暂停终止。the red team on the injection test: the fixed \(\alpha\) ignores exposure (zero point \(+23.3\) counts, 1.21\(\sigma\)); a DETNAM regex over-injects 9/102 ObsIDs by \(\times 2.3\). the audit red team on the gate code: \(\sigma_{\rm total}\) lacks V_common; the gate decision is not enforced on the target-read paths. The fix jobs P67/P69 were stopped by the pause. | 恢复后先吸收 R098/R104 修码，再跑任何上限；先读 R098、R104。After resuming, absorb the R098/R104 code fixes before running any limit; read R098, R104 first. | 已知缺陷未修Known defects, unfixed |
| 6.（关闭）档案 Chandra 叠加能否测到热 CGM？6. (Closed) Can an archival Chandra stack detect the hot CGM? | 它曾是主目标（D015/D017）。It was the main goal (D015/D017). | 关闭证据：heli474 锚重归一后最佳单元 S/N 1.17（DD0）\(\to\) 0.07（P65 包络；R103）；\(\sigma_{\rm common}\) 不可辨识（R094）；档案给不出 \(\pm 2\%\) 匹配的空白场（需 2329 块 / 626 GB，只有 937 块；R093）。D020 据此改目标为上限。Closing evidence: re-anchored to the heli474 anchor the best-cell S/N is 1.17 (DD0) \(\to\) 0.07 (P65 envelope; R103); \(\sigma_{\rm common}\) is not identifiable (R094); the archive cannot supply a matched \(\pm 2\%\) empty-field null (2329 blocks / 626 GB needed, 937 available; R093). D020 redirected the goal to an upper limit on this basis. | 不再作为探测目标；上限路线见问题 2。No longer a detection goal; the upper-limit route is problem 2. | 关闭（2026-10-22，D020）Closed (2026-10-22, D020) |
| 7.（停泊）MOS 零场偏置的成因7. (Parked) The origin of the MOS null-field offset | XMM 线移交独立项目后停泊；恢复 XMM 工作前必须先解开。Parked with the XMM line moving to a separate project; it must be solved before XMM work resumes. | 三个解释被拒绝（SP 指数、pn 泄漏、QPB 标度）；稳定天空重拟合把常规场偏置移到 \(-0.27 \pm 0.08\)，判"不确定"。Three explanations rejected (SP index, pn leak, QPB scaling); the stable-sky refit moves the normal-field offset to \(-0.27 \pm 0.08\), verdict "ambiguous". | 固定 CXB 天空或 NGC 5746 多历元检验（PI 已同意，停泊中）。The fixed-CXB sky or the NGC 5746 multi-epoch test (PI-agreed, parked). | 停泊Parked |
来源：cognition-os/02_CURRENT_QUESTION.md；cognition-os/04_DECISION_LOG.md D015、D017、D020、D021；cognition-os/07_REVISION_LOG.md R060、R067、R068、R093、R094、R095、R098、R102、R103、R104、R105；P69 REPORT.luna.md。问题 6 的关闭证据 = R103 + R094 + R093 + D020。Source: cognition-os/02_CURRENT_QUESTION.md; cognition-os/04_DECISION_LOG.md D015, D017, D020, D021; cognition-os/07_REVISION_LOG.md R060, R067, R068, R093, R094, R095, R098, R102, R103, R104, R105; the P69 REPORT.luna.md. Problem 6's closing evidence = R103 + R094 + R093 + D020.
骨架：项目各部分如何连接（暂停时）Skeleton: how the parts connect (at the pause)
来源：cognition-os/05_PROJECT_DAG.md（改画为 SVG）；cognition-os/04_DECISION_LOG.md D015/D017/D020/D021。Source: cognition-os/05_PROJECT_DAG.md (redrawn as SVG); cognition-os/04_DECISION_LOG.md D015/D017/D020/D021.
计划：恢复点（D021 给出的顺序）Plan: the resume points (in the order D021 gives)
- 恢复条件：eRASS:\(\ge 2\) 事件可用（DR3 约 2028 下半年）或 PI 另行决定。在此之前不启动任何执行任务。Resume condition: eRASS:\(\ge 2\) events become available (DR3 around the second half of 2028), or the PI decides. No executor starts before that.
- 恢复点 1——P58 驱动重启：用 sol/P58/scripts 里的 retry/skip 驱动补完半 1（35/213 已验证；部分原始数据 132 MB 可再生）。目标、输入、方法、判据见"如何接手"页步骤 2。Resume point 1 — the P58 driver restart: use the retry/skip driver in sol/P58/scripts to finish half 1 (35/213 verified; 132 MB of partial raw data, regenerable). Goal, inputs, method and criteria are step 2 on the "By hand" page.
- 恢复点 2——P70 上限机器：先吸收 R098/R104 的代码修复（曝光加权 \(\alpha\)、V_common、门强制执行），再按 P60 预注册（第四门未过时目标保持盲）给出 95% 上限，携带背景残余包络。Resume point 2 — the P70 upper-limit machinery: absorb the R098/R104 code fixes first (exposure-weighted \(\alpha\), V_common, enforced gates), then issue the 95% upper limit under the P60 pre-registration (targets stay blinded while the fourth gate fails), carrying the background-residual envelope.
- 恢复点 3——P71 Chandra 试点：30 颗 heli474 高覆盖（\(\ge 80\%\)）星系的点源辨认/掩膜试点；不把小覆盖子样本外推为全栈改善。Resume point 3 — the P71 Chandra pilot: a point-source identification/masking pilot on the 30 heli474 galaxies with \(\ge 80\%\) coverage; the small covered subsample is not extrapolated to a full-stack improvement.
- 恢复点 4——eRASS1 He & Li 复现：按 P69 顺序 1（双误差口径 + 控制带 + 随机场），再扩到 P59 的质量分箱。Resume point 4 — the eRASS1 He & Li reproduction: follow P69's order 1 (dual error conventions + control bands + random fields), then extend to P59's mass bins.
- 停泊（不随恢复自动启动）：XMM 零场 MOS 偏置（固定 CXB / 多历元检验）；原矛盾的分解实验（I001）——等 XMM 校准与独立项目。Parked (not auto-started on resume): the XMM null-field MOS offset (fixed-CXB / multi-epoch tests); the decomposition experiment of the original tension (I001) — waiting for the XMM calibration and the separate project.
来源：cognition-os/04_DECISION_LOG.md D021（恢复点原文）；08_RUNNING.md（P58/P70/P71 状态）；P69 REPORT.luna.md。Source: cognition-os/04_DECISION_LOG.md D021 (the resume points as written); 08_RUNNING.md (the P58/P70/P71 states); the P69 REPORT.luna.md.
本次更新（2026-10-22 23:43 EDT）What's new (2026-10-22 23:43 EDT)
- 项目暂停（D021，23:42 EDT）。PI 确认没有可用的 eRASS:2 光子数据；所有执行任务停止；本手册是收尾交接。恢复条件与四个恢复点见上。The project is paused (D021, 23:42 EDT). The PI confirmed no eRASS:2 photon data are available; all executors stopped; this handbook is the closing hand-off. The resume condition and four resume points are above.
- Chandra 探测目标关闭（D020）。档案 Chandra 叠加不再作为探测目标；新目标 = P58 半 1 的 95% 上限；新主线 = 复现 eROSITA 叠加并检验 Chandra 的改善点。The Chandra detection goal is closed (D020). The archival Chandra stack is no longer a detection goal; the new goal = the 95% upper limit from P58 half 1; the new main line = reproduce the eROSITA stack and test where Chandra improves it.
- 可达性证据链。P68 把预报按 heli474 锚重归一：最佳单元 S/N 1.17 \(\to\) 0.07（R103）；\(\sigma_{\rm common}\) 不可辨识（R094）；档案空白场不够 \(\pm 2\%\) 匹配（R093）。The reachability evidence chain. P68 re-anchors the forecast to heli474: best-cell S/N 1.17 \(\to\) 0.07 (R103); \(\sigma_{\rm common}\) is not identifiable (R094); the archive's empty fields cannot supply a matched \(\pm 2\%\) null (R093).
- Vrot 检验已回答（R060）。\(\Delta V = -0.307 \pm 0.178\)（按星系自助），90% CI [\(-0.622, -0.088\)]；按 D013 分层判"不确定"（负号）；红队复核 HOLDS（R067）。The Vrot test is answered (R060). \(\Delta V = -0.307 \pm 0.178\) (galaxy bootstrap), 90% CI [\(-0.622, -0.088\)]; verdict "ambiguous" under the D013 tiers (negative); the red-team review HOLDS (R067).
- eRASS1 复现可行性（P69 / R105）。He & Li 474 复现 GO（泊松 7.16\(\sigma\) / 自助 3.98\(\sigma\) vs 论文 7.4\(\sigma\)）；Zhang eRASS:4 精确复现 NO-GO（DR2 目录-only；DR3 约 2028 H2）。The eRASS1 reproduction feasibility (P69 / R105). The He & Li 474 reproduction is a GO (Poisson 7.16\(\sigma\) / bootstrap 3.98\(\sigma\) vs the paper's 7.4\(\sigma\)); the exact Zhang eRASS:4 reproduction is a NO-GO (DR2 catalogue-only; DR3 ~2028 H2).
- Chandra 背景残余是限制性问题（R095/R102）。P53 场 blank-sky \(-18.6\%\)，P58 目标场 \(+8.6\%\)/\(+12.3\%\)——场依赖 10–20%，任何常数修正都不可迁移。The Chandra background residual is the limiting problem (R095/R102). P53 fields blank-sky \(-18.6\%\), P58 target fields \(+8.6\%\)/\(+12.3\%\) — field-dependent at 10–20%; no constant correction transfers.
- 叠加代码两处缺陷确认未修（R098/R104）。曝光盲 \(\alpha\)（+23.3 计数）与 DETNAM 正则（\(\times 2.3\)）在注入检验里发现；\(\sigma_{\rm total}\) 缺 V_common 与门未强制在门代码审计里发现。恢复后先修码。Two stacking-code defects confirmed, unfixed (R098/R104). The exposure-blind \(\alpha\) (+23.3 counts) and the DETNAM regex (\(\times 2.3\)) were found by the injection test; the missing V_common in \(\sigma_{\rm total}\) and the unenforced gates were found by the gate-code audit. Fix the code first after resuming.
- P60 预注册冻结（R091）。515 对（sha256 6e47c5dc…），主几何 10–30 kpc 源环 / 150–300 kpc 背景，0.5–1.2 keV，四个强制零门；第四门未过，目标计数保持盲。The P60 pre-registration is frozen (R091). 515 pairs (sha256 6e47c5dc…), primary geometry 10–30 kpc source / 150–300 kpc background, 0.5–1.2 keV, four mandatory null gates; the fourth gate failed, so the target counts stay blinded.
数据截至Data as of ：2026-10-22 23:43 EDT（D021；cognition-os/08_RUNNING.md 2026-10-22 13:22 EDT 版 + D021 行）。: 2026-10-22 23:43 EDT (D021; cognition-os/08_RUNNING.md, the 2026-10-22 13:22 EDT revision plus the D021 row).