ParetoGTICal

ParetoGTICal 工作台ParetoGTICal workbench

评审更正（2026-10-13，Claude）Review correction (2026-10-13, Claude)本站写于 CDFS 盲测完成之前。P001 已结束（D006）：CDFS 上 E1 无法评估（主样本无零对照），E2 通过，E3 两台相机均不通过（MOS1 24/29，Z = 1.55；MOS2 27/29，Z = 1.73；增益约 0.3%）；角落精度 σ = 0.0044 / 0.0038，可用。下文凡写“CDFS 待定”处以此为准，下次刷新会更新全文。来源：cognition-os/experiments/E003_p001_cdfs_blank_field.mdThis site was written before the CDFS blind run finished. P001 is now closed (D006): on CDFS, E1 is not evaluable (no zero control in the primary set), E2 passes, and E3 fails on both cameras (MOS1 24/29, Z = 1.55; MOS2 27/29, Z = 1.73; gain about 0.3%); the corner precision is σ = 0.0044 / 0.0038, usable. Wherever the text below says "CDFS pending", this note supersedes it; the next refresh updates the pages. Source: cognition-os/experiments/E003_p001_cdfs_blank_field.md
四个问题Four questions · 目标表Objectives · 开放问题表Open problems · 计划Plan · 骨架图Skeleton · 动画Animation · 本次更新What's new · 数据截至Data as of

这是什么项目? XMM-Newton 望远镜每次观测都会被软质子 (soft proton) 耀斑污染，必须用"好时间段 (good time interval, GTI)"把耀斑时间删掉。删多删少是一对矛盾：多删 → 剩下的数据更干净但更少；少删 → 数据多但更脏。ParetoGTICal 把这个取舍当作一个帕累托问题 (Pareto problem):在单次观测内部，用保留活时间 f_T 与连续谱约减因子 R = C_all/C_K 两个坐标给每个候选 GTI 打分，并用 Al-K、Si-K 荧光线的不变性做门 (gate),防止"挑噪声"伪装成"删耀斑"。它属于父项目 DiffuseXrayCalibration （校准 XMM EPIC 弥散发射分析），但有自己的问题、自己的认知操作系统 (cognition OS) 和自己的预注册 (pre-registration) 纪律。

What is this project? Every XMM-Newton observation is polluted by soft-proton flares, and "bad" time must be removed by keeping only good time intervals (GTIs). How much to cut is a genuine conflict: cut more and the retained data are cleaner but fewer; cut less and there are more data but dirtier ones. ParetoGTICal treats this trade-off as a Pareto problem: within one observation, every candidate GTI is scored by two coordinates — retained live time f_T and the continuum reduction factor R = C_all/C_K — with an Al-K / Si-K fluorescence-line stability gate that stops "selecting noise" from masquerading as "cutting flares". It is a sub-project of DiffuseXrayCalibration (calibrating the XMM EPIC diffuse-emission analysis), with its own question, its own cognition OS and its own pre-registration discipline.

四个问题，一屏作答Four questions, answered on one screen

现在在哪？
Where does it stand?
预注册 P001 已批准、代码已冻结 (git 46d8df3),M31 MOS1 盲测已跑完：E1 通过、E2 通过、E3 失败(13/25,Z = 0.02)——率排名在相同保留率下并不比标准 2.filter 更好；角落 (corner) 在 MOS1 上太吵 （σ = 0.013 > 0.010）。CDFS 空白场盲测待定（stage-1/2 还原进行中）。
Pre-registration P001 is approved and the code frozen (git 46d8df3); the blind M31 MOS1 run is complete: E1 PASS, E2 PASS, E3 FAIL (13/25, Z = 0.02) — rate-rank at equal retention is not better than the standard 2.filter; the unexposed corner is too noisy on MOS1 (σ = 0.013 > 0.010). The CDFS blank-field blind test is pending (stage-1/2 reduction in progress).
还有什么没解决？
What is unsolved?
① CDFS 盲测（空白场对照）；② 角落到底有没有用 (T017);③ 下游成功判据（曝光损失 ⇄ 连续谱增益的兑换率）；④ 反事实耦合修正下的 Al 亏损 (−2.2σ, MOS2) 是粒子背景真变了还是未建模的系统误差。
① The CDFS blind test (blank-field control); ② whether the unexposed corner is useful at all (T017); ③ the downstream success criterion (the exchange rate between exposure loss and continuum gain); ④ whether the counterfactual coupling-corrected Al deficit (−2.2σ on MOS2) is a real particle-background change or an unmodelled systematic.
接下来做什么？
What do I do next?
等 CDFS stage-1/2 还原完成后，用冻结的代码在 M1、M2 上跑盲测并评估 E1–E6;然后定角落的最终判决；然后定义下游成功判据。命令见"如何接手"页。
When the CDFS stage-1/2 reduction finishes, run the blind test on M1 and M2 with the frozen code and evaluate E1–E6; then settle the corner verdict; then define the downstream success criterion. Commands are on the "By hand" page.

目标表（PI 工作单，D002,2026-10-13）Objectives (the PI's work order, D002, 2026-10-13)

# 目标Objective 状态Status 最好的当前数字（来源）Best current number (source) 趋势Trend
1 事实纠错认知 OSFact-correct the cognition OS DONE D002 + I002：训练/保留集（19 训练、11 保留）2026-09-14 已冻结且已消费；两个度量族 R 与 G_Al；数据是 M31 MOS2 不是 CDFS。D002 + I002: the train/holdout split (19 training, 11 holdout) was frozen and scored on 2026-09-14; two metric families R and G_Al exist; the data are M31 MOS2, not CDFS.
来源：Source: ParetoGTICal/cognition-os/04_DECISION_LOG.md (D002); insights/I002_historical_evidence_reconciliation.md
 →
2 扩展到 MOS1 + 空白场对照Extend to MOS1 + blank-field control MOS1 已完成done / CDFS 待定pending M31 MOS1 盲测：E1 通过 (0/78),E2 通过 (0 vs 0),E3 失败 (13/25,Z = 0.02),角落 σ = 0.013 → TOO_NOISY。CDFS:待定 （P001 CDFS 盲测尚未运行）。M31 MOS1 blind: E1 PASS (0/78), E2 PASS (0 vs 0), E3 FAIL (13/25, Z = 0.02), corner σ = 0.013 → TOO_NOISY. CDFS: pending (P001 CDFS blind test not yet run).
results/p001_m31_mos1/p001_endpoints_m31.json; cognition-os/experiments/E002_p001_m31_mos1.md
 →
3 定义下游成功判据Define the downstream success criterion 待定PENDING 尚无数字。它要回答：1% 的活时间值多少连续谱约减？由父项目的响应折叠弥散天空拟合给出。E3 失败后，这个问题变成"选哪个保留率",而非"选哪个时间"。No number yet. It must answer: how much continuum reduction is 1% of live time worth? Given by the parent project's response-folded diffuse-sky fit. After E3 failed, this becomes "which retention to choose", not "which time to keep".
ParetoGTICal/cognition-os/00_AI_BRIEFING.md §Next steps (D002 step 3); 04_DECISION_LOG.md D005
 →
4 开放问题：未曝光角落有没有用？Open question: is the unexposed corner useful? PARTLY TESTED MOS1 叠加 σ = 0.0133 > 0.010 → TOO_NOISY;MOS2 参考 σ = 0.0098 → 勉强 USABLE;两台相机都看不到 K/B 角落差 （−0.008 ± 0.013, −0.003 ± 0.010）。最终判决等 CDFS。MOS1 stacked σ = 0.0133 > 0.010 → TOO_NOISY; MOS2 reference σ = 0.0098 → USABLE by a hair; neither camera shows a K/B corner difference (−0.008 ± 0.013, −0.003 ± 0.010). Final verdict waits for CDFS.
results/p001_m31_mos1/p001_endpoints_m31.json (E5); 03_TRUST_TABLE.md T017
 →

开放问题表Open problems

问题Problem 为什么重要Why it matters 当前最好的假设Best current hypothesis 下一个检验Next test 所有者 / 状态Owner / status
R 筛选在等保留率下是否优于归档 2.filter?Does the R screen beat the archived 2.filter at equal retention? 这是 P001 的主要终点；若不优于，方法的全部价值就只剩"选保留率"这一层。This is P001's primary endpoint; if not, the method's whole value reduces to "choosing a retention". 不优于：两台相机 Δ 都在噪声内（MOS1 13/25,Z=0.02;MOS2 21/29,Z=1.64,都不达阈值）。Not better: both cameras have Δ within noise (MOS1 13/25, Z = 0.02; MOS2 21/29, Z = 1.64; neither meets the thresholds). CDFS 盲测（M1、M2,冻结代码，E1–E6）。The CDFS blind run (M1, M2, frozen code, E1–E6). T019 / REFUTED on M31T019 / REFUTED on M31
反事实 Al 亏损（−0.028 ± 0.013, MOS2）是真信号吗？Is the counterfactual Al deficit (−0.028 ± 0.013, MOS2) real? 若真，则 L_K = L_B 这个设计级约束本身有偏差；若假，则是未建模系统误差。它决定线门是否可信。If real, the design-level constraint L_K = L_B is itself biased; if not, it is an unmodelled systematic. It decides whether the line gate is trustworthy. MOS1 不复现 (−0.004 ± 0.019,as gated +0.019 ± 0.019),提示它是 MOS2 特有或一次涨落，且依赖注入测得的耦合 s。It does not transfer to MOS1 (−0.004 ± 0.019 counterfactual; +0.019 ± 0.019 as gated), suggesting it is MOS2-specific or a fluctuation, and that it depends on the injection-derived coupling s. CDFS 上的 E5;若 CDFS 也不出现，按 T005 关闭该问题。E5 on CDFS; if absent there too, close the issue per T005. T005, T016 / QUESTIONEDT005, T016 / QUESTIONED
角落能否预测 K/B 粒子级变化？Can the corner predict the K/B particle-level change? 角落在 FOV 外、不经望远镜聚焦，是最干净的粒子监测器；若够准，就能直接检验"Al 亏损 = 粒子变化"。The corner is outside the FOV and receives no focused soft protons — the cleanest particle monitor; if precise enough it can test "Al deficit = particle change" directly. 目前两台相机都测不出 ~1% 的 K/B 角落差；B1 曾有 +2.8% (z=1.7) 的边缘提示。Neither camera currently resolves a ~1% K/B corner difference; B1 once hinted at +2.8% (z = 1.7). CDFS E5 角落叠加 σ 是否 ≤ 0.010。Whether the CDFS E5 corner stack σ is ≤ 0.010. T017 / PARTLY TESTEDT017 / PARTLY TESTED
被移除的成分是软质子还是可变天体源？Is the removed component soft protons or variable celestial sources? 这是"删掉的东西到底是什么"的身份问题；M31 盘里有很多可变 X 射线双星，是天体源解释的真实竞争者。This is the identity question — what was actually removed; the M31 disk holds many variable X-ray binaries, a genuine celestial competitor. A1:移除的超额 ~79% 无线 (Al −5.96σ);B1:空间延展、非紧凑，更像软质子花样；但仍不能排除多个微弱可变源。A1: the removed excess is ~79% line-free (Al −5.96σ); B1: extended, not compact — closer to the soft-proton pattern; many faint variable sources are not excluded. CDFS 空白场 E6:若移除成分是软质子，耀斑 CDFS 观测的 R 增益应像 M31;若差异大，指向 M31 源。The CDFS blank field, E6: if the removed component is soft protons, R gains on flared CDFS observations should look like M31's; large differences point at M31 sources. T009, T018 / NOT ESTABLISHEDT009, T018 / NOT ESTABLISHED
来源：ParetoGTICal/cognition-os/03_TRUST_TABLE.md (T005, T009, T017–T019);04_DECISION_LOG.md (D002, D005);E002_p001_m31_mos1.md。Source: ParetoGTICal/cognition-os/03_TRUST_TABLE.md (T005, T009, T017–T019); 04_DECISION_LOG.md (D002, D005); E002_p001_m31_mos1.md.
ParetoGTICal skeleton: event list → light curve → candidate GTIs → K/B → R and line gate → null → P001 endpoints → missing downstream criterion
最新骨架。虚线框是全项目唯一缺失的下游环节：用父项目的响应折叠 (response-folded) 弥散天空拟合定义"曝光损失 ⇄ 连续谱增益"的兑换率（D002 第 3 步）。 The latest skeleton. The dashed box is the one missing downstream piece: defining the exchange rate between exposure loss and continuum gain through the parent project's response-folded diffuse-sky fit (D002 step 3).

未来 1–4 周计划（按 PI 顺序与 P001 §9）Plan for the next 1–4 weeks (PI order and P001 §9)

  1. 等 CDFS stage-1/2 还原完成（codexnim 与 Hermes 各一半，作业 2026-10-13-pareto-cdfs-mos-filter），用冻结代码在 M1、M2 上跑 scripts/m31_mos2_line_continuum_invariant.py --field cdfs --camera M1|--camera M2,再用 scripts/p001_endpoints.py 评估 E1–E6。CDFS 结果待定。Wait for the CDFS stage-1/2 reduction to finish (codexnim and Hermes each on half, job 2026-10-13-pareto-cdfs-mos-filter), run the frozen code on M1 and M2 with scripts/m31_mos2_line_continuum_invariant.py --field cdfs --camera M1|--camera M2, then evaluate E1–E6 with scripts/p001_endpoints.py. CDFS results are pending.
  2. 按 E5 的 CDFS 角落叠加 σ 给出 T017 的最终判决（可用 / 太吵）。Give the final T017 verdict (usable / too noisy) from the CDFS E5 corner stack σ.
  3. 定义下游成功判据（D002 第 3 步）：在父项目的响应折叠拟合里，量化 1% 活时间损失对应多少连续谱增益，把 f_T–R 平面变成一个可决策的标量。E3 失败后，这成为"保留率政策"问题。Define the downstream success criterion (D002 step 3): in the parent project's response-folded fit, quantify how much continuum gain 1% of live time buys, turning the f_T–R plane into a decidable scalar. After E3 failed this becomes a "retention policy" question.
来源：ParetoGTICal/cognition-os/00_AI_BRIEFING.md §Next steps;prereg/P001_mos1_and_blank_field.md §9;04_DECISION_LOG.md D005。Source: ParetoGTICal/cognition-os/00_AI_BRIEFING.md §Next steps; prereg/P001_mos1_and_blank_field.md §9; 04_DECISION_LOG.md D005.

一分钟看懂核心矛盾（动画）The core conflict in one minute (animation)

光变曲线 + 阈值滑块 → GTI、f_T、保留计数实时更新Light curve + threshold slider → GTI, f_T and retained counts update live

拖动阈值：蓝色是保留的时间 K,红色是被删的 B。f_T 是保留活时间比例；R 用玩具模型模拟"删高率时间为什么会让 R>1"。

Drag the threshold: blue is kept time K, red is rejected B. f_T is the retained live-time fraction; R uses a toy model to show why cutting high-rate time gives R > 1.

本次更新What's new

2026-10-13(第一版):本手册随 ParetoGTICal 迁移入 DXC 后建立。覆盖 D001–D005 全部决策：P001 预注册批准 + 代码冻结 (46d8df3) + 回归门通过 (26,111 个数字，最大相对差 0),M31 MOS1 盲测完成 （E002）。CDFS P001 一切结果均为"待定"。

2026-10-13 (first edition):This handbook was built after ParetoGTICal migrated into DXC. It covers decisions D001–D005: P001 pre-registration approved + code frozen (46d8df3) + regression gate passed (26,111 numbers, max relative difference 0), and the blind M31 MOS1 run (E002) recorded. All CDFS P001 results are "pending".

数据截至Data as of

2026-10-13（所读最新来源：cognition-os/04_DECISION_LOG.md D005 与 06_AGENT_LOG.md 最后一行，均为 2026-10-13;E002 实验记录同日）。

2026-10-13 (newest sources read: cognition-os/04_DECISION_LOG.md D005 and the last row of 06_AGENT_LOG.md, both 2026-10-13; the E002 experiment record is dated the same day).