这一页按时间讲项目走过的路，包括死胡同——死胡同往往比成功更能教人东西。所有版本号与结论都来自 VERSION_CHANGELOG.md 与 cognition-os。
This page tells the project's story in time order, dead ends included — dead ends often teach more than successes. All version numbers and conclusions come from VERSION_CHANGELOG.md and cognition-os.
时间线
Timeline
| 时间When | 事件Event | 教训Lesson | ||
|---|---|---|---|---|
| 2025-06 | v1–v6 探索期。确定 0.4–3.2 keV 能段（避开低能噪声与高能背景）；试过 MCMC（v2–v4，背景全自由）；v5 起固定 T=0.75 keV、Z=0.3。 | v1–v6 exploration. Fixed the 0.4–3.2 keV band (avoiding low-E noise and high-E background); tried MCMC (v2–v4, all background free); from v5 fixed T=0.75 keV, Z=0.3. | MCMC 在当时的信噪比下无法同时约束 T 与 Z。 | MCMC could not constrain T and Z simultaneously at the available S/N. |
| 2025-06–09 | v7–v12 稳定期。输出目录化；冻结背景 + 缩放因子（v11）；CXB 换成 xspegpwrlw（v12）；两步/联合拟合策略定型。 | v7–v12 stabilization. Output directories; frozen background plus a scaling factor (v11); CXB switched to xspegpwrlw (v12); the two-step/joint strategy took shape. | 先定结构、再调策略，避免一次改太多。 | Fix the structure first, then the strategy; avoid changing many things at once. |
| 2025-09–11 | v13–v15 发表冻结。CXB 固定（“临时”，等更好的背景拟合）；三种金属丰度网格；论文基线。 | v13–v15 publication freeze. CXB fixed (“temporary”, pending a better background fit); three metallicity grids; the paper baseline. | “临时”决定可能变成永久基线——记录它。 | A “temporary” decision can become the permanent baseline — record it. |
| 2026-04 | S04 SZ/X-ray 张力。X 射线导出的 SZ 比参考高约 2 倍；beta 模型数据版本混乱；MCMC 处理上限。 | S04 SZ/X-ray tension. The X-ray-derived SZ was ~2× the reference; beta-model data versions were confused; MCMC with upper-limit handling. | 数据版本对不上时先查文件时间戳，再谈物理。 | When data versions disagree, check file timestamps before physics. |
| 2025-11–2026-02 | v16–v19 温度重拟合。v16 重拟合温度 profile；v17 切换 4.background；v18 模块化绘图 + APEC 环境修复 + skyarea “修正之修正”（原版其实是对的）；v19 冻结源 SP norm=0。 | v16–v19 temperature refits. v16 refit the temperature profile; v17 switched to 4.background; v18 modular plotting + the APEC environment fix + the skyarea “correction of the correction” (the original was right); v19 froze the source SP norm to 0. | 一个“显而易见的 bug”可能是对的：先做对照再改（I037→I038）。 | An “obvious bug” may be correct: run the control before fixing (I037→I038). |
| 2026-05 | 2D 温度图第一次尝试失败。89-bin 自适应 binning 的 100 个 kT 全部 Deprecated（APEC 修复前、零流量、SNR 5–10，rstat≈15.9）；审计后确立 SNR≥25 门槛（D17/D18）。 | First 2D kT map attempt failed. All 100 kT values from the 89-bin adaptive binning Deprecated (pre-APEC-fix, zero flux, SNR 5–10, rstat≈15.9); the audit set the SNR≥25 rule (D17/D18). | 低 SNR 的拟合不是“粗略结果”，是错误结果。 | Low-SNR fits are not “rough results”; they are wrong results. |
| 2026-05–07 | 方向轴分区与扇区。0/90/180/270 象限图被判定为 debug artifact（D21）；DS9 fk5 区域格式门槛（D24）；R4–7 四扇区→八扇区（22.5° 偏置轴心，D32/D30）。 | Physical-axis regions and sectors. The 0/90/180/270 quadrant map was judged a debug artifact (D21); the DS9 fk5 region gate (D24); R4–7 four sectors → eight sectors (22.5°-offset axis-centred, D32/D30). | 几何要先过人眼审核再提取数据。 | Geometry needs human visual approval before data extraction. |
| 2026-07-13/14 | Fe-L 颜色图与 pn0 审计。Fe-L 色图只能当形态学交叉验证（I055/I060）；pn0 显示 Wind S 最软，与 MOS2 相反（2.67σ）；D40 阻止“反馈风”解释。带比值（HR→kT）方法早已被证伪（T23/T39/T40，P01）。 | Fe-L colour map and pn0 audit. The Fe-L colour map is morphology-only (I055/I060); pn0 shows Wind S softest, opposite to MOS2 (2.67σ); D40 blocks the feedback-wind interpretation. Band-ratio (HR→kT) methods had long been falsified (T23/T39/T40, P01). | 两台相机打架时，谁的背景合同更干净谁更可信——但先别下结论。 | When two cameras disagree, trust the cleaner background contract — but do not conclude yet. |
| 2026-07-15 | P0 NO-GO（D45）。统一合同下 R4–6 冷热对比 +0.1348 keV，但 18 参数 full-nuisance 95.45% 区间 [−0.069, +0.343] keV 过零；M0（等温）AICc 更低。P1 停止。 | P0 NO-GO (D45). Under the common contract the R4–6 contrast is +0.1348 keV, but the 18-parameter full-nuisance 95.45% interval [−0.069, +0.343] keV crosses zero; M0 (isothermal) has lower AICc. P1 stopped. | 点估计再漂亮也不算证据；区间过零就停。 | A pretty point estimate is not evidence; stop when the interval crosses zero. |
| 2026-10-05/06 | Chandra+XMM 联合管线停滞。8/20 单元 PASS、12 PENDING；R6–8 严格交集仅 5.2% 面积；X1/X2 给 0.109 keV 开区间，J1/J2 双分支，C1 2.04 keV。 | Chandra+XMM joint pipeline stalls. 8/20 cells PASS, 12 PENDING; the strict R6–8 intersection keeps 5.2% of the area; X1/X2 give a 0.109 keV open interval, J1/J2 bimodal, C1 2.04 keV. | “每台仪器都必须支持同一块天”会把面积砍到无法测量。 | Requiring every instrument to support the same sky cuts the area below measurability. |
| 2026-10-05 | E030–E033 温度图设计。8 个真实区域实测：盘侧 12.9 arcmin² 误差 10.0%（通过），北侧 11.6 arcmin² 32.2%（不通过）；北侧理想缩放候选 ~30.2 arcmin²；20% 验收规则 I≥30.08（lnT 近似）。 | E030–E033 temperature-map design. 8 real regions measured: disk-side 12.9 arcmin² gives 10.0% error (pass), north 11.6 arcmin² gives 32.2% (fail); the ideal-scaled north candidate is ~30.2 arcmin²; the 20% acceptance rule needs I≥30.08 (lnT approximation). | 相同 S/B 比可以有完全不同的温度精度——比例门槛不足。 | Equal S/B ratios can give very different temperature precision — ratio thresholds are not enough. |
| 2026-10-14/15 | 四问计划（Q-A…Q-D）。PI 提出：拟合可靠性、Chandra+XMM profile、13′ 过亮、2D 图。四个 job 并行派出，随后加派 GTI 审计、mask 设计、星系团检验与 Virgo 再抽谱。 | The four-question program (Q-A…Q-D). The PI asks: fitting reliability, the Chandra+XMM profile, the 13′ excess, and 2D maps. Four jobs dispatched in parallel, then the GTI audit, mask design, cluster test and the Virgo re-reduction. | 先验证方法，再产出地图（Q-A 先于 Q-D 温度图）。 | Validate the method before making the map (Q-A precedes the Q-D kT map). |
| 2026-10-15 | 13′ “亮环”破解 + 一次公开撤回。sol 读码发现 v18 保存的 flux 本身已是每 arcmin²（B020）→ 一切 flux/area 轮廓双重归一（Deprecated）。此前 02:20Z 的“非嵌套通量”论证（子区通量高于母区 → 物理不自洽）在 02:40Z 被撤回：面亮度本来就允许子区更亮，论证读错了单位。Q-C 改判为假象：裸 SB 比 1.37，冻结 kT 后 1.07，图像无超出，WARP 星系团拟合为 0（I065）。GTI 保留 2.filter；MOS1 静态硬过剩成为新开问题（I066）；mask v2 无法压残余到 <5%，必须建模（E035）；Q-B 旧管线搁置、改走逐仪器路线（D46）；Virgo 再抽谱计划（D47）。 | The 13′ “bright ring” solved + a public retraction. sol's code reading found the v18 saved flux is already per arcmin² (B020) → every flux/area profile double-normalised (Deprecated). The 02:20Z “non-nested flux” argument (a sub-region's flux exceeding its parent's → physically inconsistent) was retracted at 02:40Z: surface brightness may legitimately rise in a sub-region, and the argument had misread the units. Q-C reclassified as an artefact: bare-SB ratio 1.37, 1.07 with kT frozen, no image excess, the WARP cluster fitting to 0 (I065). Keep GTI 2.filter; the MOS1 quiescent hard excess became a new open problem (I066); mask v2 cannot push residuals below 5%, so they must be modelled (E035); the old Q-B pipeline parked for the per-instrument route (D46); the Virgo re-reduction plan (D47). | 单位错了，物理论证跟着错。在比较两个数之前先写下每个数的单位与产生代码行（B020；02:40Z 撤回）。撤回要公开、要快。 | Wrong units, wrong physics. Before comparing two numbers, write down each one's units and producing code line (B020; the 02:40Z retraction). Retract publicly and fast. |
死胡同清单（别再走进去）
Dead ends (do not walk into them again)
- HR→kT 带比值（T23 Deprecated，~0.15 keV 系统偏差；H band 92% 是背景）。Fe-L 内部硬度（0.875–1.05/0.70–0.875）是唯一有条件的交叉验证代理（I054），不能替代谱拟合。
- HR→kT band ratio (T23 Deprecated, ~0.15 keV systematic bias; the H band is 92% background). The Fe-L internal hardness (0.875–1.05/0.70–0.875) is the only conditional cross-check proxy (I054); it does not replace spectral fitting.
- pre-APEC-fix 自适应 binning 的全部 100 个 kT（D17 Deprecated）。
- All 100 pre-APEC-fix adaptive-bin kT values (D17 Deprecated).
- 0/90/180/270 象限 2D 图（T67 Deprecated：混合了盘平行与风垂直方向）。
- The 0/90/180/270 quadrant 2D map (T67 Deprecated: mixes disk-parallel and wind-perpendicular directions).
- 把点源包含的 Fe-L 色图像素当温度（T121 Deprecated）。
- Treating point-source-included Fe-L colour pixels as temperatures (T121 Deprecated).
- skyarea “修正”（I037→I038：原 v18 是对的，replot_v18_fixed_skyarea.py 已删）。
- The skyarea “fix” (I037→I038: the original v18 was right; replot_v18_fixed_skyarea.py was deleted).
- SAS 全局 fkeypar/pget 包装器（T137：污染 emchain，产生假 TUNIT；生产停止，D43/D44）。
- The global SAS fkeypar/pget wrappers (T137: they corrupted emchain and faked a TUNIT; production stopped, D43/D44).
- flux/area_arcmin2 面亮度轮廓与旧 all-method 表（B020：保存的 flux 已是每 arcmin²，再除一次 = 双重归一；all-method 表面积换算错 484 倍。均 Deprecated）。
- flux/area_arcmin2 SB profiles and the old all-method tables (B020: the saved flux is already per arcmin², so dividing again double-normalises; the all-method tables' area conversion is wrong by 484×. Both Deprecated).
- “非嵌套通量”论证（2026-10-15 02:20Z 提出、02:40Z 撤回：面亮度允许子区更亮；错误源于读错单位）。
- The “non-nested flux” argument (proposed 2026-10-15 02:20Z, retracted 02:40Z: surface brightness may rise in a sub-region; the error came from misreading the units).