diag: F 段——MeanFlow 一步采样残留噪声（开环 1/2/4/8 段与零噪声的幅度/抖动/残差-噪声相关；闭环 4 段、零噪声）；sample_k 加 steps、rollout 加 steps/z_scale；删 E 段；本机执行清单更新

ml/model/policy.py

@@ -151,8 +151,9 @@ class Policy(nn.Module):
         x = z - self.decode(m, z, torch.zeros_like(one), one)
         return x, v.expand(K), s.expand(K, -1)
 
-    def sample_k(self, feat, K, z=None):
-        """仿真 / APG 用：每个观测 K 条候选 → act [B,K,16,6], value [B], sub [B,13]"""
+    def sample_k(self, feat, K, z=None, steps=1):
+        """仿真 / APG 用：每个观测 K 条候选 → act [B,K,16,6], value [B], sub [B,13]
+        steps>1（诊断用，2026-10-19）：把 [0,1] 均分成 steps 段，x ← x − (t−r)·u(x, r, t) 逐段走；steps=1 即一步生成。"""
         mem = self.encode(feat)
         v, s = self.heads(mem)
         B = feat.shape[0]
@@ -165,7 +166,10 @@ class Policy(nn.Module):
         z = torch.randn(B * K, self.h, 6, device=feat.device, dtype=feat.dtype) if z is None else z.reshape(B * K, self.h, 6)
         m = mem.repeat_interleave(K, 0)
         one = torch.ones(B * K, device=feat.device, dtype=feat.dtype)
-        x = z - self.decode(m, z, torch.zeros_like(one), one)
+        x = z
+        for i in range(steps):
+            t, r = 1 - i / steps, 1 - (i + 1) / steps
+            x = x - (t - r) * self.decode(m, x, one * r, one * t)
         return x.view(B, K, self.h, 6), v, s