为何存在Why this project exists

一句话：XMM-Newton 的 RGS 有极高的能量分辨率，但对展源没有真正的空间分辨率；N132D 是一个几角分的展源；我们要把“位置”从“波长”里拆出来，测它的速度场。In one sentence: the XMM-Newton RGS has superb energy resolution but no true spatial resolution for extended sources; N132D spans several arcminutes; we want to pull "position" back out of "wavelength" and measure its velocity field.

1. 科学问题1. The science question

N132D 是大麦哲伦云里最亮的 X 射线超新星遗迹之一，约 2500 年。它处在“动力学年轻”的阶段：激波还在跑，等离子体还没达到碰撞电离平衡。要理解它的爆炸与演化，最直接的观测量是三维速度场：壳层朝哪个方向膨胀？快多少？各处一致吗？N132D is one of the brightest X-ray supernova remnants in the Large Magellanic Cloud, about 2500 years old. It is dynamically young: the shock is still running and the plasma has not reached collisional ionisation equilibrium. The most direct observable for understanding the explosion and its evolution is the three-dimensional velocity field: which way is the shell expanding, how fast, and is it uniform?

视向速度由谱线的多普勒频移给出。O VII 与 O VIII 窄带有最强的软 X 射线线，而且 RGS 的分辨率足以分辨几百 km/s 的移动。PI 的当前目标（D29，2026-10-19）就是：对这两条窄带建模，得到发射气体的空间分布与速度分布。Line-of-sight velocities come from Doppler shifts of spectral lines. The O VII and O VIII narrow bands hold the strongest soft X-ray lines, and the RGS resolution is enough to separate shifts of a few hundred km/s. The PI's current goal (D29, 2026-10-19) is exactly this: model the two narrow bands and obtain the emitting gas's spatial and velocity distributions.

来源：cognition-os/00_AI_BRIEFING.md（项目目标与 Current focus 一节）、cognition-os/04_DECISION_LOG.md D29、docs/science-contract.md（科学范围）。Source: cognition-os/00_AI_BRIEFING.md (the project objective and Current focus), cognition-os/04_DECISION_LOG.md D29, docs/science-contract.md (the science scope).

2. 仪器困境2. The instrument dilemma

RGS 是一台无狭缝光栅光谱仪（见“物理与统计”第 1 节）。对点源没有问题：交叉色散位置固定，波长轴干净。但对 N132D 这种几角分的展源，每个天空位置的光子都按色散定律摊到波长轴上——空间被混进了波长。一个 0.06′ 的位置错位就能伪装成 100 km/s 的速度。The RGS is a slitless grating spectrometer (see "Physics", section 1). Point sources are fine: the cross-dispersion position is fixed and the wavelength axis is clean. But for an extended source like N132D, photons from every sky position are spread onto the wavelength axis by the dispersion law: space is mixed into wavelength. A positional offset of 0.06 arcmin can masquerade as 100 km/s.

更糟：Chandra 有空间分辨率但没有 RGS 的谱分辨率；RGS 有谱分辨率但没有空间分辨率。单靠任何一台仪器都测不了展源的速度场。Worse: Chandra has the spatial resolution but not the RGS spectral resolution; the RGS has the spectral resolution but not the spatial one. Neither instrument alone can measure an extended source's velocity field.

来源：docs/architecture.md（同条带拓扑的解释限制）；cognition-os/insights/I021_cross-dispersion-offset.md（位置错位与速度的换算）。Source: docs/architecture.md (the interpretation limit of the same-row topology); cognition-os/insights/I021_cross-dispersion-offset.md (the conversion between positional offset and velocity).

3. 数据融合的思路3. The data-fusion idea

用户在 2026-10-13 澄清了项目的定位（D17）：这是数据融合——不引入第二台仪器，而是用同一台 RGS 在不同 roll 角下的互补色散投影。换 roll 只改变天空结构投影到色散几何的方式，谱本身不变；于是“空间混进波长”的方式变了，而“物理速度造成的移动”不变。多个 roll 联合拟合，就能把两者分开。The user clarified the framing on 2026-10-13 (D17): this is data fusion, not a second instrument, using the same RGS at different roll angles for complementary dispersion projections. Changing the roll changes only how sky structure projects onto the dispersion geometry, not the spectrum itself; so the way "space mixes into wavelength" changes while the "physical velocity shifts" do not. Fitting several rolls jointly separates the two.

这个思路在玩具模拟里已经量化（E013）：三个 roll 合测一个 400 km/s/′ 的梯度，精度约 5 %、方向 2°；0°+180° 最利于 \(g_x\) 与 \(v_0\)，0°+90° 利于 \(g_y\)。N132D 恰好有三个可用观测，roll 角近似 0°、90°、180°。The idea is already quantified in a toy mock (E013): three rolls together measure a 400 km/s per arcmin gradient to about 5 % with a 2-degree direction error; the 0+180 degree pair is best for \(g_x\) and \(v_0\), 0+90 for \(g_y\). N132D happens to have three usable observations at roll angles near 0, 90 and 180 degrees.

来源：cognition-os/00_AI_BRIEFING.md（定位澄清 2026-10-13）；04_DECISION_LOG.md D17；03_TRUST_TABLE.md E013 行（Trusted，玩具 mock 范围）。Source: cognition-os/00_AI_BRIEFING.md (the framing clarification of 2026-10-13); 04_DECISION_LOG.md D17; 03_TRUST_TABLE.md row E013 (Trusted, toy-mock scope).

4. 为什么要一条可复现的管线4. Why a reproducible pipeline

速度场测量的误差预算里，仪器系统误差与统计误差同量级（LSF 宽度差 2 % 就偏 24 km/s；位置差一个像素就偏 100 km/s）。所以每个数字都必须可追溯到：哪一版代码、哪一份数据、哪一个响应、哪一次拟合。S2Fitting 管线把这一切做成不可变的 run 目录加哈希清单；cognition-os 把每个结论的信任状态与证据钉在一起。这个 handbook 的每个数字也带来源路径。In the velocity error budget, instrument systematics are of the same order as the statistical errors (a 2 % LSF width error biases by 24 km/s; a one-pixel misplacement by 100 km/s). So every number must trace to a code version, a dataset, a response and a fit. The S2Fitting pipeline makes all of this an immutable run directory with a hashed manifest; cognition-os pins every claim's trust state to its evidence. Every number in this handbook carries a source path too.

来源：AGENTS.md（使命与工作规则）、docs/architecture.md（run layout 与验证）、cognition-os/03_TRUST_TABLE.md I020 与 I021 行。Source: AGENTS.md (mission and working rules), docs/architecture.md (run layout and validation), cognition-os/03_TRUST_TABLE.md rows I020 and I021.