项目现状Where things stand
数据截至：cognition-os 记录的最新时间戳 2026-10-19T11:23Z（06_AGENT_LOG.md 最后一行 2026-10-19T11:23:42Z）。本页回答三个问题：哪些已经确立、哪些还开着、每一条我们有多确定。信任标签直接取自 03_TRUST_TABLE.md；在跑的任务一律标“待定”。Data as of: the newest cognition-os timestamp, 2026-10-19T11:23Z (the last 06_AGENT_LOG.md line is 2026-10-19T11:23:42Z). This page answers three questions: what is established, what is open, and how sure we are of each. Trust labels come straight from 03_TRUST_TABLE.md; in-flight jobs are marked Pending.
本页目录On this page
1. 已确立1. Established
1.1 仪器与几何（Trusted）Instrument and geometry (Trusted)
| 结论Claim | 数值Value | 信任Trust | 来源Source |
|---|---|---|---|
| 色散偏移就是平移：交叉色散每偏一个角分，谱线移 0.1233 ÅA dispersion offset acts as a shift: each arcmin of cross-dispersion moves a line by 0.1233 Angstrom | 0.1233 Å/′ | 可信Trusted | E010 |
| 三观测平移律一致（0.1236–0.1238 Å/′）；条带接受等于各条带效率乘一个共享但不对称的形状The three observations share the shift law (0.1236-0.1238 Angstrom per arcmin); acceptance equals per-strip efficiency times a shared but asymmetric shape | 0.1236–0.1238 Å/′ | 可信Trusted | E014, I016 |
| RGS LSF 核心宽 \(\sigma\)（O VII r）；未建模展宽须低于 85 km/sRGS LSF core \(\sigma\) (O VII r); unmodelled broadening must stay below 85 km/s | 437 km/s / 85 km/s | 可信Trusted | I020 |
| 前向模型把 N132D 在交叉色散上放偏（探测器坐标）；每次观测一个自由偏移即可消除，代价不超过 0.5 %The forward model misplaces N132D in cross-dispersion (detector frame); one free offset per observation removes it at a cost below 0.5 % | \(-6.25 \pm 0.49\)″ | 可信Trusted （范围收窄）(narrowed scope) | I021 |
| 标定源的交叉色散平移随波长变化（9 到 31 Å，斜率 \(-0.21\)″/Å）；Capella 与 PKS 2155-304 十年间一致The calibrators' cross-dispersion shift varies with wavelength (9 to 31 Angstrom, slope \(-0.21\) arcsec per Angstrom); Capella and PKS 2155-304 agree across a decade | \(-0.21\)″/Å | 可信Trusted | I026 |
| 官方 angdist 展源响应与点源叠加差 1.3–2.4 %；平移仿真器 0.14–0.54 %The official angdist extended-source response deviates 1.3-2.4 % from a dense point-source superposition; the shift emulator reaches 0.14-0.54 % | 1.3–2.4 % / 0.14–0.54 % | 可信Trusted | E012 |
1.2 方法与稳健性（Trusted，mock 范围）Method and robustness (Trusted, mock scope)
| 结论Claim | 数值Value | 信任Trust | 来源Source |
|---|---|---|---|
| 正交视角恢复固定 \(3 \times 3\) 空间图所需计数更少；等总计数下，roll 角决定哪个梯度分量受益Orthogonal views need fewer counts to recover the fixed \(3 \times 3\) map; at equal total counts, the roll angle decides which gradient component benefits | RMSE 比 0.53–0.84RMSE ratio 0.53-0.84 | 可信Trusted | E013, I009 |
| 遗漏跨条带 PSF 路径会高估绝对通量（+8 到 +14 %），CStat 却几乎不变Omitting cross-strip PSF paths overestimates absolute flux (+8 to +14 %) while CStat barely moves | +8 至 +14 %+8 to +14 % | 可信Trusted | E008, I010 |
| 自由共同线宽不吸收遗漏的跨条带混合A free common line width does not absorb the omitted cross-strip mixing | \(\hat{\sigma}/\sigma_0\) 差不超过 0.01\(\hat{\sigma}/\sigma_0\) differs by at most 0.01 | 可信Trusted | E011 |
| 已知模板的玩具中，三观测合测 400 km/s/′ 梯度精度约 20 km/s/′（5 %）；运动结区是最大系统误差In a known-template toy, the three observations together measure a 400 km/s per arcmin gradient to about 20 km/s per arcmin (5 %); a moving knot is the largest systematic | 5 % / 方向 2°5 % / direction 2 deg | 可信Trusted | I017 |
| MOS 模板把 \(\sigma_v\) 压到 0、把 \(g_x\) 偏 \(-3\) 到 \(-4\) SD：真实分析需要 Chandra 级模板A MOS template pins \(\sigma_v\) at 0 and biases \(g_x\) by \(-3\) to \(-4\) SD: the real analysis needs a Chandra-grade template | \(-3\) 至 \(-4\) SD\(-3\) to \(-4\) SD | 可信Trusted | I019 |
| 壳层加结点两分量模型把结区梯度偏差从最多 10 SD 降到 0.4 SD 以内The shell-plus-knot two-component model cuts the knot's gradient bias from up to 10 SD to within 0.4 SD | 10 SD 至 0.4 SD10 SD to 0.4 SD | 可信Trusted | I018 |
来源：cognition-os/03_TRUST_TABLE.md（各行当前信任标签与范围）；范围限定词（“mock 范围”“参考观测”“玩具”）来自同一表的 Scope 列。所有 mock 结论的注入与拟合使用同一算子，不证明真实数据偏差的成因。Source: cognition-os/03_TRUST_TABLE.md (each row's current trust label and scope); the scope qualifiers ("mock scope", "reference observation", "toy") come from the same table's Scope column. All mock conclusions inject and fit with the same operator and do not prove the cause of real-data deviations.
2. 还开着2. Open
2.1 Provisional（已测但未完成全部核查）Provisional (measured but not fully verified)
| 结论Claim | 数值Value | 来源Source |
|---|---|---|
| N132D 的 O VII 气体相对 LMC 静止，速度弥散约 500 km/s；梯度只有 \(1.9\sigma\)、方向不稳健N132D's O VII gas is at rest in the LMC frame with a dispersion of about 500 km/s; the gradient is only \(1.9\sigma\) and its direction is not robust | \(v_0 = 12 \pm 21\)，\(\sigma_v = 500 \pm 27\)，\(|g| = 68 \pm 31\) km/s/′ | I029 |
| 固定分界的双成分拟合改善 \(\Delta C = -101\)；每半仍需 \(\sigma_c \approx 438\) km/s——宽度主要是内部弥散The fixed-boundary two-component fit improves \(\Delta C = -101\); each half still needs \(\sigma_c \approx 438\) km/s, so the width is mostly internal dispersion | \(\sigma_c = 438 \pm 28\)，\(v_s = -99 \pm 43\) km/s | I029 （RT21）(RT21) |
| O VIII 窄带第 1 阶段：梯度幅度约 111 km/s/′，方向与 O VII 相差 176°，弥散约 412 km/s；但泊松拟合优度不合格（p = 0.005），所以结论以速度模型为条件O VIII narrow-band stage 1: a gradient magnitude of about 111 km/s per arcmin, 176 degrees from the O VII direction, with a dispersion of about 412 km/s; but the Poisson goodness of fit fails (p = 0.005), so the conclusion is conditional on the velocity model | 111 km/s/′，412 km/s | E023 |
| 档案普查新增 11 个 N132D 附近的 RGS 观测，已全部约简（22/22 次运行，388 ks）；响应包与逐观测偏移尚待定The archive census adds 11 RGS observations near N132D, all reduced (22 of 22 runs, 388 ks); the response packs and per-observation offsets are still pending | 11 obs / 388 ks | E022 |
| 刚性配准误差沿色散方向的分量与波长零点简并，只移动 \(v_0\)（每 2″ 折合 56.6 km/s），不改变梯度；非刚性配准待定The component of a rigid registration error along dispersion is degenerate with the wavelength zero point and moves only \(v_0\) (56.6 km/s per 2 arcsec), not the gradient; non-rigid registration is pending | 56.6 km/s per 2″ | I029 （RT23）(RT23) |
| 点源响应误差对 N132D 速度的影响典型不超过 0.1 SD；E018b 讨厌参数组把它压到 0.05 SD 以内Point-source response errors affect N132D velocities by at most about 0.1 SD typically; the E018b nuisance set keeps them within 0.05 SD | 0.1 SD / 0.05 SD | I028 |
| Chandra 0.50–0.62 keV 图像可用作 O VII 模板，带不超过 0.22 SD 的污染系统误差；连续谱占比 0.13–0.16The Chandra 0.50-0.62 keV image serves as the O VII template with a contamination systematic below 0.22 SD; the continuum share is 0.13-0.16 | 0.22 SD / 0.13–0.16 | I023 （RT06、RT12 修订）(RT06, RT12 revision) |
| 交叉色散尺度 \(w \approx -3\,\%\)：不是坐标尺度误差、不是 O VII 污染、不是任何已测点源响应误差；与模板宽度简并The cross-dispersion scale \(w \approx -3\,\%\): not a coordinate-scale error, not O VII contamination, not any measured point-source response error; degenerate with the template width | \(w\) \(-0.026\) 至 \(-0.031\)\(w\) \(-0.026\) to \(-0.031\) | I025 （RT14、RT19）(RT14, RT19) |
| 条带计数里 14–38 % 来自条带外（78–95 % 来自离边 0.5′ 内）；条带外接受度整体偏 25 % 使 \(\sigma_v\) 偏 0.29–0.35 SD14-38 % of strip counts come from outside the strip (78-95 % within 0.5 arcmin of the edge); a common 25 % error in out-of-strip acceptance biases \(\sigma_v\) by 0.29-0.35 SD | 14–38 % | I027 （RT11）(RT11) |
2.2 Questioned / 撤回Questioned / withdrawn
- “一个 Chandra 模板同时服务 O VII 三重线的 r、i、f 三条线”记为 Questioned（I029，RT22，2026-10-19T09:34Z）：f 线的模板北移 2–3″ 使 C 降低 40–49，r 线拒绝同样平移；I029 的速度参数因此带 0.3–0.6 SD 的模板系统误差。天区原因还是仪器原因由 E025 检验。"One Chandra template serves the r, i and f lines of the O VII triplet together" is recorded as Questioned (I029, RT22, 2026-10-19T09:34Z): shifting the f line's template 2-3 arcsec north lowers C by 40-49, while the r line rejects the same shift; I029's velocity parameters therefore carry a 0.3-0.6 SD template systematic. Whether the cause is sky or instrumental is tested by E025.
- 同条带振幅 = 去 PSF 空间图（B002，Questioned）：响应里没有跨行混合时，这个解释不成立。拟合数值保留为历史范围。Same-row amplitudes as a PSF-deconvolved map (B002, Questioned): without cross-row mixing in the response, the interpretation does not hold. The fit values stay as history.
- I022 的“核宽 20 %”读法已撤回：那个构造其实等价于坐标缩放；纯 PSF 宽度误差（条带固定）由主 agent 精确注入补测（核加宽 20 % 以内 \(\sigma_v\) 偏差不超过 0.12 SD）。I022's "core width 20 %" reading is withdrawn: that construction is a coordinate scale in disguise; the pure PSF-width error (strips fixed) was measured by an exact main-agent injection (core broadening within 20 % biases \(\sigma_v\) by at most 0.12 SD).
- RT06 第一轮的“N VII 41 % 污染、梯度偏 0.6–0.7 SD”已撤回；用 N132D 自己的 RGS 线强核对，主要混入是 O VIII（约 21 %）。RT06's first-pass "N VII 41 % contamination, gradient biased 0.6-0.7 SD" is withdrawn; checked against N132D's own RGS line strengths, the main contaminant is O VIII (about 21 %).
- RT17 第一轮的线性偏差表符号反、放大 2 倍（把 C 统计量梯度当步长）；主 agent 改正后重算全部 16 个形状。RT17's first-pass linear-bias table had the sign reversed and doubled (it used the C-statistic gradient as the step); the main agent recomputed all 16 shapes after the correction.
来源：cognition-os/03_TRUST_TABLE.md（B002、I022 corrected、I023 RT06、E018b design RT17 行）；cognition-os/REDTEAM_QUEUE.md（各队的撤回记录）。Source: cognition-os/03_TRUST_TABLE.md (rows B002, I022 corrected, I023 RT06, E018b design RT17); cognition-os/REDTEAM_QUEUE.md (the withdrawal records of each team).
3. 在跑（待定）3. In flight (Pending)
| 任务Job | 目标Goal | 状态Status |
|---|---|---|
| E018c | O VII 第二阶段：细网格响应 + 分界自由的双成分速度分布O VII stage 2: fine-grid responses plus a two-component velocity distribution with a free split boundary | 待定Pending |
| E020 | O VIII 窄带输入包The O VIII narrow-band input pack | 待定Pending |
| E021 | RGS 档案普查与两个窄带的谱线清单The RGS archive census and the line inventory of both bands | 待定Pending |
| E024 | 11 个新观测的 O VII / O VIII 响应包Response packs for the 11 new observations | 待定Pending |
| E025 | O VII 禁戒线 f 的天空分布是否与共振线不同（天区原因还是仪器原因）Whether the O VII forbidden line f has a different sky distribution from the resonance lines (sky or instrumental cause) | 待定Pending |
| E026 | O VIII 第 2 阶段：拟合优度为何不合格O VIII stage 2: why the goodness of fit fails | 待定Pending |
| RT24 | 非刚性天区配准能否伪造梯度Whether a non-rigid sky registration can fake the gradient | 待定Pending |
| RT25 | 新观测的逐观测交叉色散偏移The per-observation cross-dispersion offsets of the new observations | 待定Pending |
| RT26 | O VIII 与 O VII 梯度为何近乎反向Why the O VIII and O VII gradients are nearly antiparallel | 待定Pending |
| handbook | 本站点的首次构建（就是你现在读的页面）This site's first build (the pages you are reading) | 待定（本地评审中）Pending (local review) |
来源：cognition-os/08_RUNNING.md 第 3 节（在飞任务、执行者与验收标准）。这些任务的结果出来后会按 OS_MAINTENANCE 分诊，任何一条在分诊前都写“待定”。Source: cognition-os/08_RUNNING.md section 3 (in-flight tasks, executors and acceptance criteria). When these jobs finish they will be triaged per OS_MAINTENANCE; until triage every one of them stays Pending.