总览 · ← 返回运行 20261003-171955-search-t2-embryo-interp-chain-12h

节点 n45

NO_CHANGE: 共有类型DETR按起源分侧移除（count/asym/preserve 8配置全≤父64.62；de_score对计数重分配结构性敏感.357→.321，代价超variogram收益）

方法说明?节点程序自带的 METHOD.md：这个程序做了什么、为什么。

来自 programs.git 90d7b3f2ea:solution/METHOD.md

调研员的计划

代码改动?这个节点的程序和父节点程序的逐行差别：绿色是新增，红色是删除。

对比：父节点版本 5684f87e61。改动的文件：solution/METHOD.md +84 −97、solution/README.md +18 −4、solution/run.py +210 −33

diff --git a/solution/METHOD.md b/solution/METHOD.mdindex 91dc1dd..8229474 100644--- a/solution/METHOD.md+++ b/solution/METHOD.md@@ -1,103 +1,90 @@-起源条件化 DETRX 漂移：单侧独有类型的检出移除按其细胞的起源阶段取 d_g^a/d_g^b（目标率/起源侧观测率）而非池化 d_g，配合 15-NN 邻域最弱置零选择（NBHDCOH）。--# 节点 42（improve，父 = 节点 40，T2:embryo:val_interp）--## 提交机制（备选，PLAN 被证否后）--父管线全部保留（mix 混抽 + procrustes3d + 阻尼 log-RMS + α=5 逐类型收敛位移 + λ=6 投影加权 +-软阈值 + β=0.2 配对收缩 + 不对称相关扩散 η_a=1.5/η_b=0.15 + 各向异性坐标整形 + 均匀幅度收缩-γ=0.25 + DETR s=2.0 + DETRX s_ext=0.7），只改表达矩阵的二值检出通道，坐标逐位不动。--### 1. DETRX-SIDE（主贡献，父 ANALYSIS next_suggestion #2「按 a-only/b-only 分别定向漂移」）--父的 DETRX 对单侧独有类型（~36% 输出细胞）用共有类型池化漂移 d_g =（Σ p_tgt·n_c + m·d̄)/(Σ k_g,c + m)-作缺失侧代理。问题：池化把 a 起源细胞（检出率 = 阶段 a 水平）和 b 起源细胞（= 阶段 b 水平）的观测-混在一个分母里，方向相互抵消——实测池化 d̄ ≈ 0.946，只移除 ~4%（3420 个 ON 项）。--修正：分母按细胞起源阶段分开累计（输出行序前半为 a 侧抽取、后半为 b 侧，`n_from_a` 切分）：--- d_g^a = (Σ p_tgt·n_c + m·d̄_a)/(Σ (k^a_g,c/n^a_c)·n_c + m) —— 目标率 / a 起源观测率-- d_g^b 同理用 b 起源细胞-- a-only 输出类型（只在早侧阶段出现的衰退谱系）用 shrink_a = max(1−d_g^a, 0)；b-only（新生谱系）用 shrink_b-- 移除数 = floor(s_ext·k_g·shrink)，s_ext = 0.7 不变（仍是「只补 70% 的缺口」的欠移除），OFF-only、-  每基因移除数按上式、伪计数平滑 m=10 不变--机制正确性（数据驱动，无外部测量）：b-only 类型的细胞直接取自晚侧阶段，其（上升）标记基因的检出在-更早的目标时刻应更低——几何插值目标 p_tgt = p_a^(1−t)·p_b^t 对上升基因低于 p_b，所以 b 起源细胞需要-向 p_tgt 回落；a-only 类型对下降基因同理。池化分母恰好把这个回落量平均掉了。实测（proxy，seed 0）：-d̄_a = 2.12（a 起源检出普遍低于目标 → a-only 几乎不移除，shrink_a 均值 0.005）、d̄_b = 0.55-（b 起源检出约为目标的 1.8 倍 → b-only 移除 48%×0.7 ≈ 34%，shrink_b 均值 0.48）；单侧细胞-1815 个（22 个类型）共移除 14257 个 ON 项（父：3420），输出 nnz 率 .0502 → .0459。--### 2. NBHDCOH（次贡献，移植自同胞节点 41 的已验证部件）--DETR/DETRX 置零选择从「该基因表达值最低的 ON 细胞」改为「15-NN 邻域（输出坐标、含自身，与-neighborhood_mmd 同一定义）内该基因均值最低的 ON 细胞」，并列时按自身值、再按行序（lexsort，确定）。-每基因移除数不变（13690 + 14257，与 lowest 逐基因一致）。生物学读法：基因在短间隔内丢失检出时，先发生在-空间环境表达最低的细胞（形态梯度驱动的区域性衰退，教科书：Gilbert, Developmental Biology, 10th ed.,-ch. 12–13；同节点 41），保持表达-位置配对。单独效果：nbhd raw .05126→.05021（local_spatial +0.51），-但与 DETRX-SIDE 叠加后 nbhd raw 回到 .05126（额外的 b-only 移除抵消了选择收益），local_spatial 组中性。--## 关键参数（提交默认，全部环境变量可复现）--- `T2_DETR_SELECT=nbhd`（`T2_NBHDCOH_K=15`）；`--ablate` 时强制 lowest-- `T2_DETRX_SIDE=1`，`T2_DETRX_STRENGTH=0.7`，`T2_DETRX_SMOOTH=10`；`--ablate` 时 side_split=False（池化）-- DETR 共有类型 s=2.0、幅度收缩 γ=0.25（均匀，父原样）、其余父参数不动-- PLAN 的 DETRSHRINK 实现保留（`T2_DETRSHRINK_ENABLE`，默认 0=关；`T2_GAMMA_BASE`/`T2_FREF`/`T2_DETRSHRINK_MODE`）--## 对照与证据--- `--ablate mechanism`（任意名）→ select=lowest、side_split=False、detrshrink 关：输出与父节点 40-  逐位一致（X/data、indices、indptr、obsm 全部 sha 级相等，已验证）。-- 机制生效证据：移除计数 3420→14257（只作用于 22 个单侧类型的 1815 个细胞，共有类型移除不变）；-  nnz .0502→.0459；四组分变化（A 半 3 种子，vs 父同日重评）：-  cell_state +2.5…+3.0（variogram raw .00954/.00986/.00934 → .00755/.00788/.00758，skill .47→.58，-  本谱系首次回到地板上；mmd_u 持平偏好）、local_spatial 中性（.05126/.05163/.05078 vs .05126/–/.05046）、-  expression_change −0.3…−1.2（de_direction .378→.365/.360/.369，de_score 在抽取噪声内）、-  shape_scale 逐位不变（坐标未动）。--## 查分记录（全部 A 半、同一评分器；父节点 40 同日重评作锚：seed0/1/2 = 63.94/63.11/63.41，均值 63.49）--PLAN 机制 DETRSHRINK（γ_g = γ_base·max(0,1−f_g/f_ref)，f_g = DETR 干跑逐基因移除比例；干跑精确：-收缩列因子 k_g≥0.75 不改 ON/OFF 模式与列内排序，移除集合与真实 DETR 逐位一致）。f_g 分布非退化：-mean .243、p90 .525、max .75；f_ref=0.5 时 γ_g mean .181、min 0、max .25，116 个支持基因中 30 个低于半缩。-6 配置（seed 0）全部 ≤ 父：--| 配置 | 榜分 | cs | ec | ls | variogram | mmd_u | nbhd | de_s | de_d |-|---|---:|---:|---:|---:|---:|---:|---:|---:|---:|-| (a) γ_base=0 | 63.58 | 53.08 | 63.01 | 60.47 | .010686 | .01122 | .05011 | .3571 | .3790 |-| (b) .25/f_ref .3 | 63.71 | 54.82 | 62.22 | 60.06 | .009756 | .01069 | .05096 | .3214 | .3777 |-| (c) .25/.5 | 63.73 | 54.93 | 62.22 | 60.01 | .009677 | .01069 | .05105 | .3214 | .3778 |-| (d) .15/.5 | 63.65 | 54.30 | 62.24 | 60.33 | .010074 | .01079 | .05039 | .3214 | .3785 |-| (e) .25/1.0 | 63.93 | 55.01 | 62.98 | 59.97 | .009604 | .01070 | .05115 | .3571 | .3777 |-| (f) inv 质量守恒反向再分配 | 62.79 | 55.38 | 59.47 | 58.54 | .009000 | .01114 | .05421 | .1786 | .3759 |-| 父重评 | 63.94 | 55.09 | 62.99 | 59.91 | .009536 | .01071 | .05126 | .3571 | .3780 |--结论：cell_state 随总收缩质量单调（γ=0 最差、反向再分配质量最大 cs 最高但 de_score 崩），-「DETR 已修正的基因少缩」的冗余假设在 s=2.0 强度下不成立——收缩与检出校正互补而非冗余；-与父节点 SPATGATE 证否的教训一致（收缩收益 ∝ 总质量）。**PLAN 证否。**--备选 NBHDCOH（单独，select=nbhd、池化漂移）：seed0 64.08（+0.14）、seed1 63.11（+0.00）；-强度扫描 seed0：x085 64.11、x095 64.13、x10 63.94、s225 63.89、s25 63.90；x085 三种子-64.11/63.12/63.53 均值 63.59（+0.10）——ls 收益真实（nbhd raw −.001）但榜分在噪声内。--提交机制（NBHDCOH + DETRX-SIDE，x07）三种子：**64.62 / 63.75 / 63.90，均值 64.09（+0.60）**，-逐种子 +0.68/+0.64/+0.49，三种子分项结构一致（cs 由 variogram 驱动 +3，ls/ec 中性偏负 <0.5）。+NO_CHANGE: 共有类型DETR按起源分侧移除（count/asym/preserve 8配置全≤父64.62；de_score对计数重分配结构性敏感.357→.321，代价超variogram收益）++## 结论++PLAN 机制 **DETR-SIDE-SHARED**（family T2EI-01）已按方案实现并查分，在替代评测+（`T2:embryo:val_interp` proxy 视图，E6.75+E8.0→E7.25，t=0.4，A 半，seed 0）上+**净负或净零**，且两种针对同一弱项（cell_state / local_spatial）的备选机制也全部+不优于父节点 42。因此本节点提交 **NO_CHANGE**：`run.py` 默认 `T2_DETR_SIDE_SHARED=0`，+权威执行输出与父节点 42 **逐位一致**（digest `fb6726bc2637ee53…`，已用 sha 级比对验证；+`--ablate mechanism` 输出亦逐位相同，故 `mechanism_active=no`）。++## 父节点基线（同尺子 A 半，seed 0，本节点自测）++我用 `--ablate`（= 池化 DETR，逐位等于父 42）在同一 A 半尺子上重测父节点，作为+所有对比的锚点（**注意**：任务书里的 parent_score 65.12 是 B 半，不能直接和 A 半比）：++- board = **64.622**；expression_change 62.69 / cell_state 58.13 / shape_scale 77.75 / local_spatial 59.91+- de_score raw .3571 (pts 7.812) · de_direction raw .3647 (7.860) · mmd_u raw .01071 (7.832) ·+  variogram raw .00755 (6.700) · neighborhood_mmd raw .05126 (14.977)+- 坐标三项逐位不变：d2_shape .00470 (8.159) · occupancy_dice .8066 (3.511) · scale_log_ratio .01620 (7.767)++## 机制确实生效（PLAN mechanism_evidence）++`T2_DETR_SIDE_SHARED=1 T2_DETR_SIDE_MODE=count s=2.0` 时，共有类型的移除按起源分侧：++- 分侧类型数 `detr_ss_n_types=10`；a 起源移除 `rem_a=2849`，b 起源移除 `rem_b=17538`，+  合计 20387，池化等价 `rem_pool=13690` → 凸性增量 +6697（+49%），其中 1683 个基因+  分侧后移除数 > 池化。增量几乎全在 **b 侧（新兴基因）**：b 起源细胞来自较晚的 E8.0，+  对上升基因检出远高于几何目标 p_tgt=p_a^.6·p_b^.4，池化时被 a 侧的低检出抵消，分侧后暴露。+- 输出 nnz 由 .04588（父）降到 .04319，说明移除确实改变了哪些细胞的哪些基因从 ON→OFF。++## 全部查分（A 半，seed 0；shape_scale 三项逐位不变=77.75，故略）++| 配置 | board | expr | cell | local | de_score raw | de_dir raw | mmd_u raw | variogram raw | nbhd raw |+|---|---|---|---|---|---|---|---|---|---|+| 父42 (=ablate) | **64.622** | 62.69 | 58.13 | 59.91 | .3571 | .3647 | .01071 | .00755 | .05126 |+| count s=2.0 (PLAN 主配置) | 64.42 | 61.94 | 58.57 | 59.43 | .3214 | .3652 | .01090 | **.00718** | .05228 |+| count s=1.25 | 64.51 | 62.05 | 58.27 | 59.95 | .3214 | .3702 | .01078 | .00742 | .05119 |+| count s=1.0 | 64.44 | 62.10 | 57.86 | 60.03 | .3214 | .3725 | .01073 | .00769 | .05101 |+| asym a=2.0 b=1.25 | 64.37 | 62.02 | 58.20 | 59.49 | .3214 | .3702 | — | .00742 | .05216 |+| asym a=2.0 b=1.0 | 64.28 | 62.08 | 57.80 | 59.50 | .3214 | .3725 | — | .00766 | .05213 |+| asym a=2.0 b=0.5 | 64.03 | 61.98 | 56.95 | 59.41 | .3214 | — | — | .00818 | .05232 |+| preserve (池化计数+分侧定向) | 64.620 | 62.75 | 58.10 | 59.87 | .3571 | .3677 | .01064 | .00761 | .05136 |++共 9 次 `vec-score`（含父基线），≥3 种幅度/解码组合，满足 §5「判无效」门槛。++## 为什么证否（根因）++1. **de_score 对「共有类型 per-gene 移除计数」的重分配结构性敏感**：任何偏离池化计数的+   分侧（count s∈{1.0,1.25,2.0}、asym b∈{0.5,1.0,1.25}）都把 de_score raw 钉在 **.3214**+   （父 .3571，−0.036，≈ expression 组 −0.75 pts）。原因：分侧的凸性增量集中移除 b 起源+   细胞上新兴基因的检出，把这些基因的 pb 压到**上升真值之下** → dp 方向对新兴基因变差 →+   de_score 命中率下降。这与父 ANALYSIS 记录的 DETRX-SIDE 代价同源（检出移除对 DE 有系统代价）。+2. **variogram 收益靠「总移除质量」，不靠「定向」**：只有把总量推到 20387（s=2.0）才把+   variogram raw 从 .00755 拉到 .00718（cell_state 组 +0.44）；s=1.0（总量 12395，接近池化）+   variogram 反而 .00769（略差）。而这个质量增量正好制造上面的 de_score 代价。+3. **唯一的 de_score 中性方案（preserve）净零**：preserve 模式保持每基因移除数=池化值+   （de_score 守住 .3571），只把移除按 excess 比例重定向到超标起源侧、侧内仍用 NBHDCOH 选择。+   结果 board 64.620 vs 父 64.622（−0.002，纯噪声）：variogram .00761、nbhd .05136 都比父略差。+   → 说明父节点池化 NBHDCOH 的**定向已近最优**，单纯改「移哪些细胞」不动分。++**净结论**：在 DETR 检出通道里，de_score 与 variogram 存在结构性张力——池化计数是 de_score+最优，任何分侧都破坏它；分侧带来的 variogram 质量收益恒小于 de_score 代价。父节点 42 的+DETR（池化 s=2.0 + NBHDCOH + DETRX-SIDE）已在此通道最优点附近，本方向是死路。++## 试过但放弃的备选（均针对 cell_state/local_spatial，均 ≤ 父）++- count 幅度扫描 s_side∈{1.0,1.25,1.5,2.0}（PLAN step 8 的 s 扫描）。+- per-side 非对称强度 asym（a 侧留满质量抓下降基因的 variogram，b 侧调轻以护 de_score）——+  de_score 仍钉死 .3214，证明代价来自「计数模式改变」本身而非 b 侧幅度大小。+- preserve（计数守恒、仅重定向选择）——唯一 de_score 中性者，但净零。+- **未做**（PLAN step 8 明确排除，且属参数调优非本节点范围）：γ（RECAL 幅度收缩）重扫、+  s_ext/伪计数 m 重扫、occupancy_dice 坐标探针（那是 shape_scale 家族 T2EI-06，非本节点弱项）。  ## 验证过 / 未验证 -- 验证：A 半三种子（上表）；ablate 与父逐位一致；seed 0 重跑逐位确定；vec-check 通过；-  运行时 25.6 s、峰值内存 0.62 GB（限额 30 min / 28 GB）。-- 未验证：B 半与正式分（查分额度已用尽，20/20）；真实 final 视图（括号相邻、共有类型更多、-  单侧细胞占比更小 → DETRX-SIDE 的贡献预期变小，但方向由程序从输入现场计算，无写死；-  obs_sum_a/b 退化时自动回退池化 d_g）；单输入回退路径未变（不走 DETR）；s_ext 在 side 条件下的-  再扫描（0.85/1.0）未测——额度限制，x07 是三种子验证过的点。-- 合规：全部量从 view 输入现场计算；未使用保留阶段/基因型的任何测量值；未读 view 以外路径。+- 已验证：seed 0 确定性（跑两次 digest 相同）；`--ablate` 与默认(off) 输出逐位=父 42+  （digest `fb6726bc…`，nnz .04588≈父报告 .0459）；`vec-check` 在 seed 0/1 均 ok；+  运行 30s / 峰值内存 0.60GB（限 30min / 28GB）。单输入阶段视图走早期 return（无 bracket），+  分侧代码不触发，与父一致。分侧只在 do_split（两侧各≥3 细胞）时启用，小类型回退池化。+- 未验证：B 半与 seed 1/2 的正式分（A 半 seed 0 已一致 ≤ 父，且根因是结构性的，+  换半/换 seed 不会改变 de_score↔variogram 张力的符号，故未耗额度复测）。+- 视图无关：新增代码只读 n_from_a、out_labels、表达与 p_tgt（均由输入现场算），+  不读 board/mode/绝对时间、不写死视图名/细胞数；默认关闭时行为=父 42（父已过伪装视图检查）。  ## 知识来源 -- 检出率随时间的一阶（几何/log-加性）衰减模型：父节点 DETR 文档同源（几何插值 p_a^(1−t)·p_b^t）；-  DETRX-SIDE 只是把同一目标的分母换成正确的起源侧基线，纯数据计算，无外部数值。-- 区域性发育不同步 / 形态梯度驱动的表达衰退（NBHDCOH 的选择依据）：Gilbert, Developmental Biology,-  10th ed., ch. 12–13（教科书，只用定性程序，不用任何保留阶段测量值）；部件本身移植自同胞节点 41-  （已验证 +0.63）。+本机制是纯统计再校准（检出率几何插值 + 按起源分侧计数），不涉及保留阶段/保留基因型的+测量值，也不引入新的教科书发育事件；动机来自父节点 42 的 DETRX-SIDE 教训「做目标率/观测率+校准时，分母必须与被校正细胞的来源同分布」（本节点内实验记录，非外部数据）。PLAN.sources 为空。diff --git a/solution/README.md b/solution/README.mdindex 6104ff4..304ba32 100644--- a/solution/README.md+++ b/solution/README.md@@ -1,8 +1,21 @@-# 节点 42：DETRX-SIDE 起源条件化漂移 + NBHDCOH 邻域置零选择（T2:embryo:val_interp）+# 节点 45：NO_CHANGE — 共有类型 DETR 按起源分侧移除（DETR-SIDE-SHARED）已证否++PLAN 机制 DETR-SIDE-SHARED（把父 42 的 DETRX-SIDE「按起源分侧」思想推广到**共有类型**的 DETR+移除计数）已实现并在替代评测 A 半 seed 0 上查分：count 模式 s∈{1.0,1.25,2.0}、per-side 非对称+asym b∈{0.5,1.0,1.25}、以及计数守恒的 preserve 备选，共 8 个配置全部 ≤ 父 42（64.62）。根因：+de_score 对共有类型 per-gene 移除计数的任何重分配结构性敏感（.3571→.3214，≈ expression 组 −0.75），+恒超过分侧带来的 variogram 收益（cell_state 组最多 +0.44）；唯一 de_score 中性的 preserve 模式净零+（board 64.620 vs 64.622）。**提交默认 `T2_DETR_SIDE_SHARED=0`，权威输出与父节点 42 逐位一致**+（digest `fb6726bc2637ee53`），`--ablate` 亦相同（mechanism_active=no）。分侧代码保留、默认关闭。+详见 METHOD.md（含完整查分表与根因分析）。++---++# 父节点 42：DETRX-SIDE 起源条件化漂移 + NBHDCOH 邻域置零选择（T2:embryo:val_interp）  父节点 40 管线全部保留（mix 混抽 + procrustes3d + 阻尼 log-RMS + α=5 逐类型收敛位移 + λ=6 投影加权 + 软阈值 + β=0.2 配对收缩 + 不对称相关扩散 + 各向异性坐标整形 + 均匀幅度收缩 γ=0.25 + DETR s=2.0 +-DETRX s_ext=0.7）。本节点改动（只动表达的二值检出通道，坐标逐位不动）：+DETRX s_ext=0.7）。节点 42 改动（只动表达的二值检出通道，坐标逐位不动）：  1. **DETRX-SIDE**（提交主机制）：单侧独有类型（~36% 输出细胞）的检出漂移代理从池化 d_g 改为按细胞    起源阶段条件化——a-only 类型用 d_g^a = 目标率/a 起源观测率，b-only 用 d_g^b。池化把两侧方向抵消@@ -14,5 +27,6 @@ DETRX s_ext=0.7）。本节点改动（只动表达的二值检出通道，坐    全低于父 63.94）：cell_state 随总收缩质量单调，收缩与检出校正互补而非冗余。代码保留    （T2_DETRSHRINK_ENABLE，默认关）。 -`--ablate mechanism` → select=lowest + 池化漂移 + 均匀收缩，输出与父节点 40 逐位一致（已验证）。-默认运行 seed 0 逐位确定；25.6 s / 0.62 GB。详见 METHOD.md。+节点 45 的 `--ablate mechanism`（等价默认 off）→ 共有类型回到池化 DETR，其余（NBHDCOH 选择、+DETRX-SIDE）不变，输出与父节点 42 逐位一致（已用 digest 验证）。默认运行 seed 0 逐位确定；+30 s / 0.60 GB。diff --git a/solution/run.py b/solution/run.pyindex f638aab..ba8d17d 100644--- a/solution/run.py+++ b/solution/run.py@@ -8,6 +8,26 @@ exp(log r_a + SCALE_DAMP·t·Δlog r), and draws cells stratified by type: round(t·n) from the later stage, the rest from the earlier one. Coordinates travel with the cells. n is log-linear in t, clipped to the board range. +This node (45, family T2EI-01): PLAN mechanism DETR-SIDE-SHARED — extend node+42's origin-conditioned drift to the SHARED-type DETR removal counts (split each+shared type's output cells by origin stage a/b, per-side target+floor(s·max(0, k_side − p_tgt·n_side))) — was implemented and FALSIFIED on the+proxy A half (seed 0). Every config lands ≤ parent 42 (64.622): count mode+s∈{1.0,1.25,2.0} → 64.44/64.51/64.42; per-side asymmetric b∈{0.5,1.0,1.25} →+64.03/64.28/64.37; the count-preserving "preserve" backup → 64.620. Root cause:+de_score is pinned at raw .3214 for ANY redistribution of the shared-type+per-gene counts (vs .3571 pooled) — the convexity increment over-removes+emerging-gene detections on b-origin cells, pushing their pb below the rising+truth; that −0.036 de_score (−0.75 expression pts) always exceeds the variogram+gain (+0.44 cell_state pts max, and only at the highest removal mass). Only the+exact-pooled-count "preserve" mode is de_score-neutral, and it is net-neutral+(targeting alone moves nothing → parent's pooled NBHDCOH is already near-optimal+). SUBMITTED = NO_CHANGE: default T2_DETR_SIDE_SHARED=0, so the authoritative+output is bit-for-bit parent node 42 (digest fb6726bc…, verified); --ablate+mechanism gives the same output (mechanism_active=no). The side-split code+(count / preserve / asym modes) is retained behind the env flag for the record.+See METHOD.md for the full grid and the de_score↔variogram tension analysis.+ This node (42, family T2EI-01): PLAN mechanism DETRSHRINK — per-gene amplitude shrink γ_g = γ_base·max(0, 1 − f_g/f_ref) conditioned on the DETR removal fraction f_g — was implemented with an exact removal-count dry run and@@ -497,6 +517,77 @@ DETRX_SMOOTH = float(os.environ.get("T2_DETRX_SMOOTH", "10")) # off (--ablate or T2_DETRX_SIDE=0) reproduces the pooled-drift parent # bit-for-bit. DETRX_SIDE = os.environ.get("T2_DETRX_SIDE", "1") == "1"+# DETR-SIDE-SHARED (node 45, family T2EI-01, PLAN mechanism): origin-split+# detection correction for the SHARED types. DETRX-SIDE proved that a pooled+# denominator cancels the drift signal (a-origin cells sit above the geometric+# target for falling genes, b-origin cells below it; pooled, the excesses+# offset). The same cancellation exists inside the shared-type DETR counts:+# rem_g = floor(s·max(0, k_g − p_tgt·n_c)) is computed on the pooled k_g/n_c,+# so a gene whose a-origin detection overshoots the target while its b-origin+# detection undershoots it gets a small pooled excess — and the NBHDCOH+# selection then zeroes cells from BOTH origins, including b-origin cells that+# are already at/below target. DETR-SIDE-SHARED splits every shared type's+# output cells by origin stage (a-drawn rows first, rows < n_from_a) and+# computes per-side removal targets+#   r_g^a = floor(s·max(0, k_g^a − p_tgt·n^a)),  r_g^b = floor(s·max(0, k_g^b − p_tgt·n^b)),+# with the same s = DETR_STRENGTH and p_tgt = p_a^(1−t)·p_b^t. By convexity of+# max(0,·) the total r_g^a + r_g^b ≥ pooled rem_g: the increment is exactly the+# per-side excess that pooling used to cancel. Selection stays NBHDCOH, applied+# within each origin group (15-NN neighborhood mean of gene g on the output+# coordinates, lowest first; counts per side per gene as above). Types with+# < 3 cells on either side fall back to the pooled count (same as parent).+# Single-input views never reach this code (no bracket → early return).+# NODE 45 OUTCOME = NO_CHANGE: the mechanism (and its count-preserve / per-side+# asymmetric backups) was FALSIFIED on the proxy A half, seed 0 — every config+# lands ≤ parent (count s∈{1.0,1.25,2.0} → 64.44/64.51/64.42; asym b∈{0.5,1.0,+# 1.25} → 64.03/64.28/64.37; preserve → 64.620; parent 64.622). Root cause:+# de_score is pinned at raw .3214 for ANY redistribution of the shared-type+# per-gene removal counts (vs .3571 pooled) — the convexity increment+# over-removes emerging-gene detections on b-origin cells, pushing their pb+# below the rising truth; that −0.036 de_score (−0.75 expression pts) always+# exceeds the variogram gain (+0.44 cell pts max). Only the exact-pooled-count+# "preserve" mode is de_score-neutral, and it is net-neutral (targeting alone+# moves nothing). Default OFF → normal run == parent node 42 bit-for-bit.+# See METHOD.md for the full grid.+# on (T2_DETR_SIDE_SHARED=1) with mode "count" reproduces the PLAN literal;+# off (--ablate mechanism or default) → parent node 42 bit-for-bit.+DETR_SIDE_SHARED = os.environ.get("T2_DETR_SIDE_SHARED", "0") == "1"+# DETR_SIDE_STRENGTH (node 45): removal strength for the SPLIT (per-origin)+# branch ONLY. The pooled DETR_STRENGTH = 2.0 was tuned on the pooled count,+# where a-origin excess and b-origin deficit CANCEL — the pooled rem is much+# smaller than the true per-side excess. Exposing that excess at s = 2.0+# over-corrects (seed 0 A half: rem 13690 → 20387, de_score raw .357 → .321,+# nbhd .0513 → .0523, board −0.20 despite variogram improving .00755 → .00718).+# A lower split strength dials the per-origin removal back toward the pooled+# total while keeping the better TARGETING (each side corrected to p_tgt on its+# own denominator). The pooled branch and the --ablate control keep+# DETR_STRENGTH = 2.0, so this parameter does not affect parent reproduction.+DETR_SIDE_STRENGTH = float(os.environ.get("T2_DETR_SIDE_STRENGTH", str(DETR_STRENGTH)))+# DETR_SIDE_MODE (node 45): how the origin split sets the per-gene removal+# COUNT for shared types.+#   "count"    — PLAN literal: rem_a+rem_b = floor(s_side·exc_a)+floor(s_side·exc_b)+#                ≥ pooled (convexity). Improves variogram via extra MASS but+#                changes rem_per_gene → de_score cost (seed 0 A half −0.036 raw),+#                net board ≤ parent at every s_side tested. (parent default)+#   "preserve" — backup mechanism: keep the pooled per-gene count rem_g EXACTLY+#                (de_score-neutral), but ALLOCATE those rem_g removals to the+#                a/b origin sides proportional to each side's excess over p_tgt+#                and select within side by NBHDCOH. Redirects zeroing onto the+#                truly-overshooting cells (parent's pooled NBHDCOH may zero the+#                below-target side) → covariance/coherence gain without touching+#                per-gene pb magnitude.+DETR_SIDE_MODE = os.environ.get("T2_DETR_SIDE_MODE", "count")+# DETR_SIDE_STRENGTH_A / _B (node 45): optional per-origin strengths for the+# "count" mode. The convexity increment is dominated by the b side (emerging+# genes on b-origin cells; rem_b ≈ 6× rem_a at s = 2.0), and over-removing those+# emerging detections pushes their pb BELOW the rising truth → the de_score cost.+# The a side (falling genes) removal LOWERS pb toward the falling truth, which is+# de_score-neutral-or-helpful. Setting _B < _A keeps the a-side variogram mass+# while softening the b-side de_score hit. Default None → both use side_strength.+_ssa = os.environ.get("T2_DETR_SIDE_STRENGTH_A")+_ssb = os.environ.get("T2_DETR_SIDE_STRENGTH_B")+DETR_SIDE_STRENGTH_A = float(_ssa) if _ssa is not None else None+DETR_SIDE_STRENGTH_B = float(_ssb) if _ssb is not None else None # DETRSHRINK (node 42, family T2EI-01, PLAN mechanism): DETR-conditional # per-gene amplitude shrink. The node-38 uniform shrink (γ = 0.25) was # calibrated on a base WITHOUT detection correction; DETR+DETRX (node 40)@@ -743,7 +834,10 @@ def recalibrate(expr, pb_a, pb_b, t, gamma, cap, support, min_dp, mode, smax, def detection_recalibrate(expr, out_labels, stage_a, stage_b, t, strength,                           min_cells, select, rng,                           detr_x=False, x_strength=0.35, x_smooth=10.0,-                          coords=None, nbhd_k=15, n_from_a=None, side_split=False):+                          coords=None, nbhd_k=15, n_from_a=None, side_split=False,+                          origin_split_shared=False, side_strength=None,+                          side_mode="count", side_strength_a=None,+                          side_strength_b=None):     """DETR: per shared cell type, remove excess gene detections toward the     geometric interpolation p_tgt = p_a^(1−t)·p_b^t of the bracket detection     rates. OFF-only: zero the weakest (lowest-value) ON entries, deterministic;@@ -775,6 +869,10 @@ def detection_recalibrate(expr, out_labels, stage_a, stage_b, t, strength,     n_touched = 0     n_zeroed = 0     n_types = 0+    # DETR-SIDE-SHARED (node 45) diagnostics: split types, per-side removals+    # vs the pooled equivalent (mechanism_evidence #1/#2 of the PLAN).+    ss_info = {"n_types": 0, "rem_a": 0, "rem_b": 0, "rem_pool": 0,+               "n_genes_gt_pool": 0}     G = expr.shape[1]     # Per-gene bookkeeping for DETRSHRINK (node 42): ON totals before any     # removal and per-gene removed-entry counts (shared DETR + one-sided DETRX).@@ -838,24 +936,84 @@ def detection_recalibrate(expr, out_labels, stage_a, stage_b, t, strength,                     obs_sum_b += (nz[n_a_c:].sum(axis=0).astype(np.float64) / n_b_c) * n_c         rem = np.floor(strength * np.maximum(k - pt * n_c, 0.0)).astype(np.int64)         rem = np.minimum(rem, k.astype(np.int64))-        rem_per_gene += rem         rows_c = np.where(mo)[0]-        for g in np.where(rem > 0)[0]:-            r = int(rem[g])-            on_local = np.where(nz[:, g])[0]-            if select == "random":-                pick = rng.choice(on_local, size=r, replace=False)-            elif nnmean is not None:-                # NBHDCOH: primary key = neighborhood mean of gene g, tie-break-                # by own value then row order (lexsort: last key is primary).-                pick = on_local[np.lexsort(-                    (Xc[on_local, g], nnmean[rows_c[on_local], g]))[:r]]+        # DETR-SIDE-SHARED (node 45): per-origin removal targets for shared+        # types. rows_c is ascending, a-drawn rows are the global rows+        # < n_from_a, so the first n_a_c rows of Xc are the a-origin cells.+        n_a_c = int((rows_c < n_from_a).sum()) if n_from_a is not None else 0+        n_b_c = n_c - n_a_c+        do_split = bool(origin_split_shared and n_from_a is not None+                        and n_a_c >= 3 and n_b_c >= 3)+        if do_split:+            s_side = strength if side_strength is None else float(side_strength)+            k_a = nz[:n_a_c].sum(axis=0).astype(np.float64)+            k_b = nz[n_a_c:].sum(axis=0).astype(np.float64)+            exc_a = np.maximum(k_a - pt * n_a_c, 0.0)+            exc_b = np.maximum(k_b - pt * n_b_c, 0.0)+            if side_mode == "preserve":+                # Count-preserving origin-priority: total per gene = pooled rem,+                # allocated to sides by excess share (de_score-neutral).+                exc_tot = exc_a + exc_b+                ka = k_a.astype(np.int64)+                kb = k_b.astype(np.int64)+                with np.errstate(invalid="ignore", divide="ignore"):+                    share_a = np.where(exc_tot > 0, exc_a / np.maximum(exc_tot, 1e-12),+                                       k_a / np.maximum(k, 1e-12))+                rem_a = np.rint(rem.astype(np.float64) * share_a).astype(np.int64)+                rem_a = np.clip(rem_a, 0, ka)+                rem_b = np.clip(rem - rem_a, 0, kb)+                # give back any cells the b clip dropped to the a side (count keep)+                deficit = rem - rem_a - rem_b+                if deficit.any():+                    room = ka - rem_a+                    add = np.minimum(deficit, room)+                    rem_a = rem_a + add+                rem_split = rem_a + rem_b+                rem_per_gene += rem_split+            else:+                sa = s_side if side_strength_a is None else float(side_strength_a)+                sb = s_side if side_strength_b is None else float(side_strength_b)+                rem_a = np.minimum(+                    np.floor(sa * exc_a).astype(np.int64), k_a.astype(np.int64))+                rem_b = np.minimum(+                    np.floor(sb * exc_b).astype(np.int64), k_b.astype(np.int64))+                rem_split = rem_a + rem_b+                rem_per_gene += rem_split+            ss_info["n_types"] += 1+            ss_info["rem_a"] += int(rem_a.sum())+            ss_info["rem_b"] += int(rem_b.sum())+            ss_info["rem_pool"] += int(rem.sum())+            ss_info["n_genes_gt_pool"] += int((rem_split > rem).sum())+        else:+            rem_split = None+            rem_per_gene += rem+        for g in np.where((rem_split if do_split else rem) > 0)[0]:+            if do_split:+                groups = ((int(rem_a[g]), 0, n_a_c), (int(rem_b[g]), n_a_c, n_c))             else:-                vals = Xc[on_local, g]-                pick = on_local[np.argsort(vals, kind="stable")[:r]]-            expr[rows_c[pick], g] = 0.0+                groups = ((int(rem[g]), 0, n_c),)+            for r, lo, hi in groups:+                if r <= 0:+                    continue+                sub = nz[lo:hi, g]+                on_local = np.where(sub)[0]+                if on_local.size == 0:+                    continue+                r = min(r, int(on_local.size))+                if select == "random":+                    pick = rng.choice(on_local, size=r, replace=False)+                elif nnmean is not None:+                    # NBHDCOH: primary key = neighborhood mean of gene g, tie-break+                    # by own value then row order (lexsort: last key is primary).+                    pick = on_local[np.lexsort(+                        (Xc[lo:hi][on_local, g],+                         nnmean[rows_c[lo:hi][on_local], g]))[:r]]+                else:+                    vals = Xc[lo:hi][on_local, g]+                    pick = on_local[np.argsort(vals, kind="stable")[:r]]+                expr[rows_c[lo + pick], g] = 0.0+                n_zeroed += r             n_touched += 1-            n_zeroed += r     x_types = 0     x_cells = 0     x_zeroed = 0@@ -926,6 +1084,11 @@ def detection_recalibrate(expr, out_labels, stage_a, stage_b, t, strength,     info.update(detrx_n_types=x_types, detrx_n_cells=x_cells, detrx_n_zeroed=x_zeroed)     info["detr_rem_per_gene"] = rem_per_gene     info["detr_on_per_gene"] = on_per_gene+    info.update(detr_side_shared=bool(origin_split_shared and n_from_a is not None),+                detr_ss_n_types=ss_info["n_types"],+                detr_ss_rem_a=ss_info["rem_a"], detr_ss_rem_b=ss_info["rem_b"],+                detr_ss_rem_pool=ss_info["rem_pool"],+                detr_ss_n_genes_gt_pool=ss_info["n_genes_gt_pool"])     info.update(detr_n_types=n_types, detr_n_genes_touched=n_touched,                 detr_n_entries_zeroed=n_zeroed,                 detr_nnz_after=float((expr != 0).mean()),@@ -1936,20 +2099,23 @@ def main() -> None:     parser.add_argument("--out", required=True)     parser.add_argument("--seed", type=int, default=0)     parser.add_argument("--ablate", default=None,-                        help="mechanism-off control (node 42): 'mechanism' (or any name) disables "-                             "the NBHDCOH detection-removal selection (falls back to lowest-own-"-                             "value) and the DETR-conditional per-gene shrink (falls back to node "-                             "38's uniform γ = 0.25), reproducing parent node 40 output bit-for-bit")+                        help="mechanism-off control (node 45): 'mechanism' (or any name) sets "+                             "origin_split_shared=False, so the SHARED-type DETR removal counts are "+                             "pooled (parent behaviour) instead of split by cell origin stage. "+                             "NBHDCOH selection and DETRX-SIDE (parent node 42 mechanisms) stay ON, "+                             "so the ablated output reproduces parent node 42 bit-for-bit")     args = parser.parse_args()-    # Mechanism-off control (node 42): NBHDCOH (+DETRSHRINK, falsified and off-    # by default). DETR+DETRX themselves (node 40's mechanism, inherited) stay-    # ON in both runs with unchanged strengths and counts; the ablated run-    # selects removals by lowest own value and applies the parent's uniform-    # shrink, so the ablated output is bit-for-bit node 40.-    detr_select = "lowest" if args.ablate else DETR_SELECT-    detrshrink_on = DETRSHRINK_ENABLE and not args.ablate-    spatgate_on = SPATGATE_ENABLE and not args.ablate-    sidecluster = 0 if args.ablate else SIDECLUSTER+    # Mechanism-off control (node 45): DETR-SIDE-SHARED (origin-split shared-type+    # removal counts). This is the ONLY mechanism this node adds on top of parent+    # node 42. NBHDCOH selection (detr_select="nbhd") and DETRX-SIDE (side_split)+    # are inherited from node 42 and stay ON in BOTH runs, so the --ablate run+    # reproduces node 42's normal output bit-for-bit. DETRSHRINK / SPATGATE /+    # SIDECLUSTER are off by default (inherited from node 42) and untouched here.+    detr_select = DETR_SELECT+    detrshrink_on = DETRSHRINK_ENABLE+    spatgate_on = SPATGATE_ENABLE+    sidecluster = SIDECLUSTER+    origin_split_shared = DETR_SIDE_SHARED and not args.ablate     proj_eta = None     proj_eta2 = None @@ -2032,9 +2198,12 @@ def main() -> None:     # DETR (node 40 mechanism, inherited): detection-rate geometric     # interpolation, applied AFTER the amplitude shrink. NBHDCOH (node 42     # backup) changes only WHICH ON entries are zeroed (selection by lowest-    # 15-NN neighborhood mean instead of lowest own value); counts, strengths,-    # the uniform shrink and coordinates are unchanged. --ablate restores-    # select="lowest" + uniform shrink → node 40 bit-for-bit.+    # DETR (node 40 mechanism, inherited): detection-rate geometric+    # interpolation, applied AFTER the amplitude shrink. NBHDCOH (node 42)+    # changes only WHICH ON entries are zeroed; DETRX-SIDE (node 42) conditions+    # the one-sided drift on origin. DETR-SIDE-SHARED (node 45, this node's+    # mechanism) splits the SHARED-type removal counts by origin stage.+    # --ablate turns off ONLY origin_split_shared → node 42 bit-for-bit.     detr_on = DETR_ENABLE or DETRX_ENABLE     if detr_on:         out_labels = info.get("out_labels")@@ -2045,7 +2214,12 @@ def main() -> None:                 detr_x=DETRX_ENABLE, x_strength=DETRX_STRENGTH, x_smooth=DETRX_SMOOTH,                 coords=coords, nbhd_k=NBHDCOH_K,                 n_from_a=int(info.get("n_from_a", 0)) or None,-                side_split=(DETRX_SIDE and not args.ablate))+                side_split=DETRX_SIDE,+                origin_split_shared=origin_split_shared,+                side_strength=DETR_SIDE_STRENGTH,+                side_mode=DETR_SIDE_MODE,+                side_strength_a=DETR_SIDE_STRENGTH_A,+                side_strength_b=DETR_SIDE_STRENGTH_B)             info.update(detr_info)     sc_info = {"sidecluster": sidecluster}     if sidecluster and expr.shape[0]:@@ -2117,6 +2291,9 @@ def main() -> None:                                           "detrx_d_min", "detrx_d_max",                                           "detrx_side", "detrx_d_a_mean", "detrx_d_b_mean",                                           "detrx_shrink_a_mean", "detrx_shrink_b_mean",+                                          "detr_side_shared", "detr_ss_n_types",+                                          "detr_ss_rem_a", "detr_ss_rem_b",+                                          "detr_ss_rem_pool", "detr_ss_n_genes_gt_pool",                                          "sidecluster", "sidecluster_n_a", "sidecluster_applied")}     print(json.dumps({"bracket": [a["stage"], b["stage"]], **keep}, default=float), file=sys.stderr)     write_t2(args.out, expr, coords, genes, seed=args.seed)

调研来源?调研员查到并用到的知识条目和文献检索结果（只列标题和编号）。

用到的知识库条目

分析结果?分析员写的 ANALYSIS.json：改了什么、各组分数怎么变、假设是否成立、经验和下一步建议。

对话摘要?每个角色和大模型对话的统计：轮数、工具调用、用时、token 数和最后的回答摘录；原始记录只给路径。

只给统计和最后回答的摘录；完整对话请到原始记录位置里列出的文件看。

分析员

ANALYSIS.json written

工程师

## 总结

**做了什么**：实现 PLAN 机制 DETR-SIDE-SHARED（共有类型 DETR 移除计数按细胞起源阶段 a/b 分侧），并针对同一弱项（cell_state/local_spatial）实现两种备选：per-side 非对称强度、计数守恒的 preserve（池化计数+分侧定向）。共 8 个配置全部查分（A 半 seed 0，9 次查询）。

**查分结果**（父 42 A 半基线 64.622，同尺子自测）：count s∈{1.0,1.25,2.0}→64.44/64.51/64.42；asym b∈{0.5,1.0,1.25}→64.03/64.28/64.37；preserve→64.620。**全部 ≤ 父**。机制确实生效（rem_a=2849/rem_b=17538 vs 池化 13690，nnz .0459→.0432），但根因是结构性的：**de_score 对共有类型 per-gene 移除计数的任何重分配都钉死在 .3214（父 .3571）**——分侧的凸性增量集中过移除 b 起源新兴基因检出、把 pb 压到上升真值之下，−0.75 expression 恒超 variogram 最多 +0.44 cell_state；唯一 de_score 中性的 preserve 净零。提交 **NO_CHANGE**：默认输出与父 42 逐位一致（digest `fb6726bc…` 已验证），`vec-check` 通过，30s/0.60GB，METHOD.md 首行 `NO_CHANGE:` + 完整查分表与根因分析。

**下一步最值得试**：本谱系 DETR 检出通道已在最优点（de_score↔variogram 张力），建议换通道——(1) 父 next_suggestion #4 的 **occupancy_dice 坐标探针**（唯一低于地板的排名指标 3.53<4.17，与表达正交、无 de_score 风险，潜在 +0.6）；(2) 直接以 **node 43（rank3 65.50，NBHDCOH-SHARED）** 为父，而非 node 42。

调研员

expression_change（62.47）因DETRX-SIDE额外~11k移除已下滑−0.97，需要更精准的移除来避免进一步损伤。ANALYSIS next_suggestion #3明确建议此方向。",
  "approach": "在父节点42全部管线上（mix+procrustes3d+阻尼log-RMS+α=5收敛位移+λ=6投影加权+软阈值+β=0.2配对收缩+不对称相关扩散η_a=1.5/η_b=0.15+各向异性整形+均匀幅度收缩γ=0.25+DETR s=2.0+DETRX-SIDE s_ext=0.7+NBHDCOH k=15），只改共有类型DETR的移除计数与选择：\n1. 按n_from_a把每个共有类型的输出细胞分为a起源与b起源两组。\n2. 对每个基因g，分别计算a起源检出率p_a_obs=k_g^a/n^a和b起源检出率p_b_obs=k_g^b/n^b（伪计数平滑m=10，与DETRX-SIDE一致）。\n3. 移除目标：a起源侧移除数r_g^a=floor(s·max(0, k_g^a − p_tgt·n^a))，b起源侧r_g^b=floor(s·max(0, k_g^b − p_tgt·n^b))，s=2.0不变，p_tgt=p_a^(1−t)·p_b^t不变。总移除数r_g^a+r_g^b≥池化版（max(0,·)的凸性），增量来自两侧各自的超额检出不再被对侧抵消。\n4. 移除选择在各自起源组内用NBHDCOH（15-NN邻域该基因均值最低），与现有选择逻辑一致。\n5. 护栏：de_direction raw≥0.3845（父值），若低于则机制证否、回退。同时监测de_score raw（目标≥0.3448，父值）。\n6. 用vec-score A半3种子（0/1/2）评估；先seed 0快筛，若榜分≥65.12且de_direction≥0.3845，再跑seed 1/2。\n7. 单输入阶段退路：若只有一个输入阶段，无a/b之分，origin_split退化为False（池化DETR），与父行为一致。\n8. 参数搜索：主配置s=2.0/m=10；若de_direction下降，试s∈{1.5,2.0}×m∈{3,10}（4配置）。不扫γ（γ重扫是参数调优，不在本节点范围内）。",
  "expected_groups": ["cell_state", "local_spatial"],
  "risks": "1. 分侧后总移除数增加（凸性），可能重演DETRX-SIDE的de_score代价——用de_direction≥0.3845硬护栏尽早发现（seed 0即可判断）。2. 共有类型两侧检出率差异可能小于单侧类型，分侧收益在噪声内（<1分）——若variogram raw改善<0.0003且榜分<65.12+1，判定无效。3. NBHDCOH在更小候选集（单侧组内）上选择可能不够稳定——监测nbhd raw，若>0.0500则告警。4. 运行时间：cKDTree已在NBHDCOH中建好，分侧只增加计数循环，预期<10s。",
  "family_id": "T2EI-01",
  "mechanism": "共有类型DETR的检出移除按细胞起源阶段（a/b）分侧计算目标与选择，避免两侧检出率在池化分母中相互抵消，使移除更精准地指向真正超出几何插值目标的细胞。",
  "vs_constant_shift": "不是对表达值加常数或乘系数；它改变的是二值检出模式（哪些细胞的哪些基因从ON变OFF），且移除目标按每个细胞的起源阶段条件化——a起源和b起源细胞对同一基因的移除数不同，取决于各自的检出率与几何插值目标的差距。",
  "mechanism_evidence": "1. 逐基因报告r_g^a与r_g^b：若分侧有效，应看到下降基因的r_g^a>r_g^b（a起源检出偏高需多移除）、上升基因反之；若r_g^a≈r_g^b≈池化/2，说明两侧检出率接近、分侧无额外信息。2. 总移除数对比（池化vs分侧）。3. variogram raw变化（目标≤0.0073）与mmd_u（目标≤0.01008）。4. de_score/de_direction是否守住护栏。5. nbhd raw（目标≤0.04988）。",
  "mechanism_off_control": "--ablate mechanism设origin_split_shared=False（共有类型回到池化DETR），其余不变，输出应与父节点42逐位一致（sha级验证）。预期差别：关闭后variogram raw回到~0.0074，若分侧有效则开启时更低。",
  "sources": []
}

审查员

REVIEW.json written