History — past lives and this life前世今生

2026-05-30 — the RGS finding (where it started)2026-05-30 —— RGS 发现（起点）

The M31Center project's fiducial RGS fit found a hot component that the paper's frozen values did not predict: kT=0.74 keV, O=1.93, Fe=0.89 (wilm abundances), with a very small vapec_norm (2.15e-4 vs the paper's 1.84e-3) and a harder power law (PhoIndex 1.67 vs 2.19). The worry: the power law absorbs most of the flux, so the 0.7 keV component might be a degenerate solution rather than real M31 emission. That worry is the project's charter question, and it is still open.

M31Center 项目的 fiducial RGS 拟合发现了一个论文冻结值无法预测的热成分：kT=0.74 keV，O=1.93，Fe=0.89（wilm 丰度），vapec_norm 非常小（2.15e-4，而论文为 1.84e-3），幂律更硬（PhoIndex 1.67 vs 2.19）。担忧在于：幂律吸收了大部分通量，所以 0.7 keV 成分可能是退化解而非真实的 M31 发射。这个担忧就是项目的立项问题，至今仍未解决。

2026-06-25 — the pivot to a region system2026-06-25 —— 转向区域系统

The project pivoted from "fit 0.7 keV per region" to "design a region system that separates three physical origins": old stars, star-formation feedback/ISM, and diffuse hot halo/CGM. The E\(\times\)P\(\times\)A hierarchy (elliptical annuli \(\times\) physical masks \(\times\) azimuthal subdivision) was defined so each region serves at least one of three orthogonal comparisons: fixed radius / different physics, fixed physics / different radius, fixed both / different azimuth.

项目从“在每个区域拟合 0.7 keV”转向“设计一个能分开三个物理起源的区域系统”：旧恒星、恒星形成反馈/ISM、弥散热晕/CGM。E\(\times\)P\(\times\)A 层级（椭圆环带 \(\times\) 物理掩护 \(\times\) 方位细分）被定义成每个区域至少服务三种正交对比之一：固定半径/不同物理、固定物理/不同半径、固定两者/不同方位。

2026-07 — the extraction campaigns (and their dead ends)2026-07 —— 提取运动（及其死胡同）

MOS-only variable-W1 annuli were extracted incrementally: a=0–4' (E009, 84/84 gates), a=4–12' (E011, 86/86), a=12–20' (E014, 86/86), a=60–80' in one pointing (E018). Several things failed and are worth knowing: the 0402560101 C-mode attempt was rejected (out-of-detmap/off-axis response warnings, MOS1 zero counts in the first two annuli — E020); a custom detector-plane clipping route was rejected by the user and must never be used (E021); legacy XRISM ROTBOX seeds were deprecated (B001); the 2026-07-17 direct-FK5 8' aggregate files were rejected because they excluded the outer ellipse and included the inner one despite valid-looking filenames (B018). On 2026-07-19 the user decided the MOS 1.4–2.0 keV interval must be included in the current campaign (it had been excluded in earlier pilots).

MOS-only 变 W1 环带被逐步提取：a=0–4'（E009，84/84 门控）、a=4–12'（E011，86/86）、a=12–20'（E014，86/86）、单指向的 a=60–80'（E018）。有几件事失败，值得记住：0402560101 的 C-mode 尝试被拒（响应有 detmap 外/离轴警告，MOS1 在前两个环带计数为零——E020）；自定义探测器平面裁剪路线被用户否决，绝不可再用（E021）；旧 XRISM ROTBOX 种子被弃用（B001）；2026-07-17 生成的直接 FK5 8' 聚合文件被拒，因为尽管文件名看似有效，它们却排除外椭圆、包含内椭圆（B018）。2026-07-19 用户决定当前运动必须包含 MOS 1.4–2.0 keV 区间（早期试点曾将其排除）。

2026-08 — the v1.6.1 extraction campaign2026-08 —— v1.6.1 提取运动

The v1.6.1 ObsID-centric control plane and the ten-region one-click controller were built (E028, E031) and then the generic campaign template (E032). The campaign went through several failed runs before it worked: run2 stopped all 33 ObsIDs at the 3B preflight (an all-EPIC detect child for a MOS-only root); run3 proved that changing the 3B config to all-EPIC is invalid; run4 stopped on a stable-publication auditor expecting 300–7200 eV names while the active config generated 500–2000 eV names; run5's pilot failed because nominal circular overlap did not translate into usable per-camera CCD/count/response support. The adopted successor semantics are task-level fail-soft: each MOS1/MOS2 \(\times\) region task is independently validated, and a failed MOS1 task does not block MOS2 or later regions. Run6 then completed: 33 ObsIDs (6 full, 27 partial), 490 complete eight-file MOS bundles, 3,920 FITS, 252 ObsID-region combinations.

v1.6.1 ObsID 中心控制平面和十区域一键控制器被建立（E028、E031），随后是通用运动模板（E032）。在成功之前，运动经历了几次失败运行：run2 在 3B 预检处拦停了全部 33 个 ObsID（MOS-only 根却发出 all-EPIC detect 子进程）；run3 证明把 3B 配置改为 all-EPIC 是无效的；run4 因稳定发布审计器期望 300–7200 eV 名称而停止，而活动配置生成 500–2000 eV 名称；run5 的试点失败，因为名义圆形重叠没有转化为可用的逐相机 CCD/计数/响应支持。被采纳的后继语义是任务级 fail-soft：每个 MOS1/MOS2 \(\times\) 区域任务独立验证，失败的 MOS1 任务不阻塞 MOS2 或后续区域。随后 run6 完成：33 个 ObsID（6 个完整，27 个部分），490 个完整八文件 MOS 包，3,920 个 FITS，252 个 ObsID-区域组合。

2026-09 — the annulus profile, the halo refit, and the killed template fit2026-09 —— 环带剖面、晕场重拟合与被终止的模板拟合

The 8' annulus profile was released (2026-09-07/09): SrcApec kT 0.71–0.90 keV over 9 retained annuli, with obs/pred rising 0.38 \(\to\) 1.64. E033 refit the halo fields for a 0.7 keV component (2026-09-09). Then the E034 old/young template fit was launched (2026-09-07): old_only completed (\(\chi^2=11771.817\), dof 7550) and young_only completed (12541.776), but mixed ran for hours at ~57.8 GiB RSS, drove host load average to ~45–57, and was killed by the PI on 2026-09-17 to restore shared-machine availability (~55 GiB released). No number from the mixed mode exists. Science then stood still from 2026-09-17 to 2026-11-01.

8' 环带剖面发布（2026-09-07/09）：9 个保留环带上 SrcApec kT 0.71–0.90 keV，obs/pred 从 0.38 \(\to\) 1.64 上升。E033 重拟合晕场以寻找 0.7 keV 成分（2026-09-09）。随后 E034 old/young 模板拟合启动（2026-09-07）：old_only 完成（\(\chi^2=11771.817\)，dof 7550），young_only 完成（12541.776），但 mixed 运行数小时、约 57.8 GiB RSS，把主机负载均值推到约 45–57，并于 2026-09-17 由 PI 终止以恢复共享机器可用性（释放约 55 GiB）。mixed 模式没有任何数字。之后科学工作从 2026-09-17 停滞到 2026-11-01。

2026-10 — delivery and language2026-10 —— 交付与语言

October was about making the results readable and reachable rather than new: the SrcApec matrix bilingual mirror (2026-09-04), the M31 MOS atlas's English half repaired across two subtrees (B023/B024, 2026-10-29/30), and the cross-project Pages directory audited repeatedly with no-op verdicts (B020/B021/B022). The PI's global rule of 2026-10-30 set English as the default language for every web page with a Chinese switch.

十月的重点不是新结果，而是让结果可读可达：SrcApec 矩阵双语镜像（2026-09-04）、M31 MOS 图谱的英文半边在两个子树中修复（B023/B024，2026-10-29/30）、跨项目 Pages 目录反复审计且结论均为无操作（B020/B021/B022）。PI 2026-10-30 的全局规则规定每个网页默认英文，配中文切换。

2026-11-01 — the full heartbeat, and the withdrawals2026-11-01 —— 完整 heartbeat 与撤回

At 10:43 EST the PI put the project under the full heartbeat (D19) and the idle SrcApec monitor cron was removed (it had been snapshotting an unchanged finished campaign every 30 min). Science restarted on free arms. In the same day the project audited itself hard and withdrew its own claims: R003/R004 withdrew the E033 halo branch-B upper limit and branch-C "detection" (the branches are not nested: B froze the sky, C added 84 soft-proton parameters); R008 questioned the "not purely stellar" \(\chi^2\) reading (conditional errors, strong SrcApec–foreground degeneracy); R013 questioned the E034 "old beats young" result once the WISE sky pedestal was found (B026); R015 found the South/SE baseline offset not robust; R016 contradicted the frozen disk nH=0.03 (MW-only foreground is about twice that). V1's evaluation gate failed at 955.007 vs 944.492 and was resumed with CORRECTION_04. V11 reported STATUS: BLOCKED. The handbook first build (this site) was launched at 20:10 EST.

10:43 EST，PI 将项目置于完整 heartbeat 下（D19），并移除了空转的 SrcApec 监控 cron（它每 30 分钟快照一个早已完成且无变化的运动）。科学工作在免费机臂上重启。同一天，项目严格自审并撤回了自己的声明：R003/R004 撤回 E033 晕场分支 B 上限和分支 C“检出”（分支不嵌套：B 冻结天空，C 增加 84 个软质子参数）；R008 质疑“非纯恒星”\(\chi^2\) 读数（条件性误差，SrcApec–前景强退化）；R013 在发现 WISE 天空底座后质疑 E034“old 胜 young”结果（B026）；R015 发现 South/SE 基准偏移不稳健；R016 否定了冻结的圆盘 nH=0.03（仅 MW 前景约为其两倍）。V1 的评估门控在 955.007 vs 944.492 处失败，并以 CORRECTION_04 恢复。V11 报告 STATUS: BLOCKED。手册首次构建（本网站）于 20:10 EST 启动。

2026-11-02 — three numbers, one foreground2026-11-02 —— 三个数字，一个前景

The day after the heartbeat started, the project ran its red-team queue and literature arms hard, and the picture changed. The hard/soft test (E045) said the inner disk needs no extra soft emission and is hard-rich (H/S 1.9–3.0), while the outer disk cannot be decided by hard/soft at all. The soft floor that the disk fits leave behind, \(s_0 = 0.31\pm0.02\times10^{-15}\), turned out to equal the halo-field 0.7 keV term they had frozen at zero (R024) — and the twenty halo-field norms turned out to be one level, \(0.97\pm0.12\times10^{-7}\) at \(8\sigma\), flat in distance from M31 (R027). Put next to the literature prediction of unresolved Milky Way coronal stars (0.72–1.05e-7) and HaloSat's published Milky Way hot component toward M31 (3.28e-16 against our 2.9e-16), three independent numbers landed on the same value. The dispatcher wrote that up as I021 and, at 03:33 EST, labelled it Provisional rather than a hypothesis (R032).

heartbeat 启动的第二天，项目狠跑了红队队列和文献机臂，图景随之改变。硬/软测试（E045）表明内圆盘 不需要额外软辐射且硬质丰富（H/S 1.9–3.0），而外圆盘根本无法用硬/软判定。圆盘拟合留下的软底板 \(s_0 = 0.31\pm0.02\times10^{-15}\)，原来等于它们冻结为零的晕场 0.7 keV 项（R024）——而二十个晕场 norm 原来是一个统一水平：\(0.97\pm0.12\times10^{-7}\)，\(8\sigma\)，与到 M31 的距离无关（R027）。 把它与未解析银河系日冕恒星的文献预测（0.72–1.05e-7）以及 HaloSat 发表的指向 M31 的银河系热成分 （3.28e-16 vs 我们的 2.9e-16）放在一起，三个独立的数字落在同一个值上。调度器把它写成 I021， 并在 03:33 EST 把它标为临时性而非假设（R032）。

Two things narrowed in the same window, and both were the project's own doing. The WISE sky pedestal was re-tested (R022): it is real, but it does not cause the old/young degeneracy; the narrow W4dust range does, so the earlier claim was re-labelled rather than repeated. And the project's own accepted fits were re-checked at their saved best point: eight of ten had stopped away from their \(\chi^2\) minimum, mostly along the plasma temperature (B027) — the numbers survive as Gaussian approximations within \(1\text{--}2\sigma\), but the refits now start from two temperatures. The PSF-spill explanation for the inner hard excess was proposed (R028), then attacked: the wings are not in the data (R033). Two decisions changed how extraction works: every early extraction is now made twice, natively and with fast_regions (D23), and the native side uses the current production pipeline end to end from the ODF (D24).

同一窗口内有两件事被收窄，而且都是项目自己做的。WISE 天空底座被重新检验（R022）：它是真的，但 不导致 old/young 退化；真正的原因是 W4dust 范围太窄，所以早先的声明被重新标注而非重复陈述。 而项目自己已接受的拟合也被在保存的最佳点上复查：十个中有八个停在其 \(\chi^2\) 最小值之外，主要沿等离子体 温度方向（B027）——这些数字作为 \(1\text{--}2\sigma\) 内的高斯近似仍然成立，但重拟合现在从两个温度起始。 内硬超额的 PSF 溢出解释被提出（R028），随后又被攻击：数据中并没有这些翼（R033）。两项决定改变了提取方式： 每次早期提取现在都做两遍，原生加 fast_regions（D23），且原生方使用当前生产管线从 ODF 到端运行（D24）。

Lessons教训

• A valid test must be nested: freezing part of the model makes the comparison conditional, not a test (R003/R004).有效检验必须是嵌套的：冻结模型的一部分会让比较变成条件性的，而不是检验（R003/R004）。
• Errors computed with the nuisance parameters fixed are not the errors of the quantity you are quoting (R008).在 nuisance 参数固定下计算的误差，不是你所引用量的误差（R008）。
• Before a survey mosaic becomes a template, compare its on-source level with an off-source sky level; a template whose range across windows is a few per cent is a pedestal, not a tracer (B026).在调查马赛克成为模板之前，先比较其源上水平与场外天空水平；跨窗口变化只有几个百分点的模板是底座，不是示踪物（B026）。
• A filename that looks right is not a geometry: the 8' aggregates excluded the outer ellipse and included the inner one (B018).看起来正确的文件名不等于正确的几何：8' 聚合文件排除了外椭圆却包含了内椭圆（B018）。
• One pointing does not make an annulus, and nominal overlap does not make a spectrum (E020).一个指向不构成一个环带，名义重叠也不构成一个谱（E020）。
• When a host rule changes, re-check the running briefs that use that host the same cycle (B025).当主机规则改变时，同一周期内要复查使用该主机的在跑简报（B025）。
• A foreground you switch off is not a foreground you have measured: when the Milky Way 0.7 keV term is frozen at zero, its photons are fitted as if they came from M31 (R024, I021).你关掉的前景不是你测量的前景：当银河系 0.7 keV 项被冻结为零时，它的光子被当作来自 M31 来拟合（R024，I021）。
• Re-check an accepted fit at its saved best point before using its errors: eight of ten had stopped short of the \(\chi^2\) minimum (B027).在使用已接受拟合的误差之前，先在它保存的最佳点上复查：十个中有八个提前停止，未到 \(\chi^2\) 最小值（B027）。