最新状态Live status

• 2026-10-16T15:1xZ 红队 RT-13 复核了 L005 的上限：计算本身和输入都成立，判定阈值 2.71 在“参数不能为负”的边界情形下用法正确；报出的上限偏保守（换温度换算或更稳的插值会再收紧 2–10%）。但上限主要靠一个视场（0800730601）：去掉它，上限放宽约一半，核半径 8 kpc、β 0.4 处与 Zhang 等 2024 名义中值的差距从 5.2 倍降到 3.3 倍。对外措辞改为“全样本 ≳4 倍；去掉约束最强的视场仍 ≥3 倍”。（可信度：Provisional (threshold uncalibrated)） 来源：cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (RT-13 triage)Red team RT-13 re-checked the L005 limits: the arithmetic and inputs hold, and the 2.71 threshold is used correctly for a parameter bounded at zero; the quoted limits are conservative (the temperature used for the flux conversion or a better-behaved interpolation would tighten them by 2–10%). But the limit rests mainly on one field (0800730601): without it the limit loosens by about half, and at core radius 8 kpc and β 0.4 the margin below the nominal median of Zhang et al. 2024 falls from 5.2 to 3.3 times. Public wording: "≳4 times for the full sample; still ≥3 times without the most constraining field". (trust: Provisional (threshold uncalibrated)) Source: cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (RT-13 triage)
• 2026-10-16T14:2xZ 外场亮度随观测年份下降的检验（RT-09-T2）：下降的是 M31 晕分量和本地热泡分量，同时一个 1.2–1.4 keV 的自由高斯残差线随年份上升；软质子和宇宙 X 射线背景在误差放大后无趋势，残差平坦，数据处理版本一致。红队 RT-14 指出：晚期外场比早期暗约一半，而官方公布的 MOS 有效面积校准误差只有几个百分点，且污染随时间的变化已由 arfgen 计入，所以单靠“探测器老化”解释不了这么大的差别；那条高斯线并非已认定的仪器荧光线，它与背景谱在同一能段的计数率一起变化（MOS2），而外场使用的校准文件版本已无法追溯。结论：原因未定（未修正的响应误差、粒子背景扣除、真实天空结构都可能），不再写“怀疑探测器老化”。下一步：固定校准文件、分早晚期重新提取，并检查背景谱和每块 CCD 的随时间变化（RT-09-T3）。在此之前，外/内比和基于外场的极限仍标为“受观测年代混淆”。（可信度：Questioned (epoch-confounded)） 来源：cognition-os/experiments/E025_outer_classB_fits.md (RT-09-T2 section; RT-14 triage)Test of the outer-field brightness falling with observing year (RT-09-T2): the M31 halo and Local Hot Bubble components fall, while a free 1.2–1.4 keV Gaussian residual line rises with year; soft protons and the cosmic X-ray background show no trend after error scaling, residuals are flat, and the processing version is uniform. Red team RT-14 points out that the late outer fields are about half as bright as the early ones, whereas the published MOS effective-area calibration error is a few percent and arfgen already includes the time-dependent contamination, so 'detector ageing' alone cannot explain a difference this large; the Gaussian is not an identified instrumental fluorescence line, it varies together with the background-spectrum count rate in the same band (MOS2), and the calibration-file versions used for the outer fields can no longer be traced. Conclusion: the cause is unresolved (an uncorrected response error, particle-background subtraction, or real sky structure are all possible); 'detector ageing suspected' is withdrawn. Next: re-extract early and late epochs with pinned calibration files, and check the background spectra and each CCD over time (RT-09-T3). Until then the outer/inner ratio and outer-based limits stay marked as epoch-confounded. (trust: Questioned (epoch-confounded)) Source: cognition-os/experiments/E025_outer_classB_fits.md (RT-09-T2 section; RT-14 triage)
• 2026-10-16T13:2xZ 更密的似然网格（14 个视场 × 117 点，0 个拟合失败）已验收。用它重算 M31 集中型 β 模型晕的上限：核半径 ≤ 8 kpc 时，20 kpc 处的面亮度上限为 0.05–0.16（单位 1e-15 erg/cm²/s/arcmin²，取决于 β），比 Zhang 等 2024 年叠加样本的名义中值 0.83 低 5 倍以上；对外仍写“≳4 倍”。线性与样条插值之差为 18–23%，取较宽松的样条值。判定阈值 Δstat=2.71 尚未用能谱模拟校准（待 E028）。（可信度：Provisional (threshold uncalibrated)） 来源：cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (A1d accepted and A2 baseline on A1d)The denser likelihood grid (14 fields × 117 points, 0 failed fits) is accepted. Recomputed with it, the upper limit on a concentrated M31 beta-model halo (core radius ≤ 8 kpc) at 20 kpc is 0.05–0.16 (units 1e-15 erg/cm²/s/arcmin², depending on β), more than 5 times below the nominal median 0.83 of the Zhang et al. 2024 stack; the public wording stays "≳4 times". Linear and spline interpolation differ by 18–23%; the looser spline value is quoted. The Δstat = 2.71 threshold is not yet calibrated with spectral simulations (pending E028). (trust: Provisional (threshold uncalibrated)) Source: cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (A1d accepted and A2 baseline on A1d)
• 2026-10-16T10:0xZ 红队 RT-11：E026 在 Virgo 上重提取 28 个外场的流程有三处错误，会让每个场在跑了几个小时之后才在审计这一步失败；CCF 定标也没有锁定到 v19 的版本。修正单已备好，修正通过之前不启动批处理。Virgo 的 ssh 已于约 08:5xZ 恢复。（可信度：Blocking） 来源：cognition-os/REDTEAM_QUEUE.md (RT-11); .agent/jobs/2026-10-15-e026-virgo-reextract-stage0/CORRECTION_01_RT11.mdRed team RT-11: the E026 Virgo re-extraction runner for the 28 outer fields has three bugs that would make every field fail at the audit step after hours of processing, and its CCF calibration is not pinned to the v19 set. A correction list is ready; no batch starts before it passes. Virgo ssh returned at about 08:5xZ. (trust: Blocking) Source: cognition-os/REDTEAM_QUEUE.md (RT-11); .agent/jobs/2026-10-15-e026-virgo-reextract-stage0/CORRECTION_01_RT11.md
• 2026-10-16T09:0xZ 外场（35–255 kpc）流量随观测年代变低：2009 年及以前 10 个场 1.05 ± 0.08，2010 年以后 3 个场 0.55 ± 0.09，而内场全部是 2017–2018 年。两组差异按置换检验约 2σ（p = 0.027）。2010 年后的偏低不是太阳周期 SWCX（太阳活动相近，流量差约 2 倍）；2001–2009 年那段下降可能部分是 SWCX。因此「外场与内场在约 10–15% 内一致」及依赖外场的大核心上限，现在标为受观测年代混杂。（可信度：Questioned (epoch-confounded)） 来源：cognition-os/experiments/E025_outer_classB_fits.md (RT-09, RT-09-T1, RT-12 sections)The outer-field (35–255 kpc) flux is lower in later observations: 10 fields up to 2009 give 1.05 ± 0.08, 3 fields from 2010 on give 0.55 ± 0.09, while the inner fields are all from 2017–2018. The split is about 2σ by permutation (p = 0.027). The post-2010 deficit is not solar-cycle SWCX (similar solar activity, about 2× different flux); the 2001–2009 decline may partly be SWCX. So 'outer and inner agree within about 10–15%' and the large-core limits that use the outer fields are now marked epoch-confounded. (trust: Questioned (epoch-confounded)) Source: cognition-os/experiments/E025_outer_classB_fits.md (RT-09, RT-09-T1, RT-12 sections)
• 2026-10-16T09:0xZ E027（ParetoGTICal 的 GTI 策略在全部 31 个观测上的能谱检验）在取数之前修订了判据：场间 τ 只作描述（功效太低），策略之间比较各自独有的时间，以软质子归一化和不依赖模型的软波段面亮度为主要指标。非 v19 观测的覆盖收益约为 4 个有用、5 个边缘指向，而不是约 11 个。203 个 GTI 文件已通过核验。（可信度：Pre-registered (amendment v2)） 来源：cognition-os/experiments/E027_gti_policy_screen.md (RT-10 section)E027 (spectral test of the ParetoGTICal GTI policies on all 31 observations) revised its metrics before any data: the field-to-field τ is descriptive only (too little power); policies are compared on the time each keeps alone; the soft-proton norm and a model-independent soft-band surface brightness are the primary channels. The coverage gain from non-v19 observations is about 4 useful and 5 marginal pointings, not about 11. The 203 GTI files passed verification. (trust: Pre-registered (amendment v2)) Source: cognition-os/experiments/E027_gti_policy_screen.md (RT-10 section)
• 2026-10-16T07:2xZ 红队 RT-08：A2 上限的 toy-MC 校准只在「整条统计曲线平移」这一简化模型下成立；换成高斯抽样，覆盖率降到 0.78–0.88。各场曲线明显不对称，这个简化不成立。A1d 之后用模拟能谱重新校准（E028）。Δstat = 2.71 这个阈值本身在理想高斯情形下仍正确。（可信度：Questioned） 来源：cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (RT-08 section)Red team RT-08: the A2 toy-MC calibration holds only under the simplified 'shift the whole statistic curve' model; with a Gaussian sampler the coverage drops to 0.78–0.88. The per-field curves are strongly asymmetric, so the simplification fails. After A1d the limits will be recalibrated with simulated spectra (E028). The Δstat = 2.71 rule itself is still right in the ideal Gaussian case. (trust: Questioned) Source: cognition-os/experiments/E013_joint_spatial_profile_likelihood.md (RT-08 section)