exp+fix: commit2 增益重扫（gain2.2 到达 50.8%，n=512）+ 浏览器安全层三点修复（限位夹紧/部分前缀/离散认证）

- gain 1.8/2.2/2.5/2.8 = 50.4/52.0/49.6/47.7%（n=256），2.2 最佳；n=512 复核 50.8%/1.4% 碰撞
- 积分反馈与反馈隔次复用均为负结论（−4.0/−5.1 pp）
- JS：tick/候选投影夹 LIM（原越限后 inBox 永久拒绝→刹停）；safety_partial 移植；segCert 开关（默认严格）
- bench nc10 --seg 0：刹车率 79.5→48.1%、planP50 56.1→33.5ms、ok 0→5/10

ml/eval/eval_cl.py

@@ -42,6 +42,7 @@ def main():
     ap.add_argument('--ki', type=float, default=0.0)
     ap.add_argument('--kdecay', type=float, default=0.9)
     ap.add_argument('--kimax', type=float, default=0.5)
+    ap.add_argument('--fb_every', type=int, default=1)
     ap.add_argument('--select', default='consist')
     ap.add_argument('--replan', default='r3')
     ap.add_argument('--safety', type=int, default=1, help='1=安全层+部分前缀；0=关')
@@ -62,7 +63,8 @@ def main():
     t0 = time.perf_counter()
     r = run(model, kin, tasks, a.device, K=a.K, safety=bool(a.safety), seed=a.seed, max_sec=a.max_sec,
             safety_partial=(a.commit if (a.safety and a.partial) else 0), commit_n=a.commit,
-            select=a.select, replan=a.replan, fb_gain=a.fb, fb_ki=a.ki, fb_decay=a.kdecay, fb_imax=a.kimax)
+            select=a.select, replan=a.replan, fb_gain=a.fb, fb_ki=a.ki, fb_decay=a.kdecay, fb_imax=a.kimax,
+            fb_every=a.fb_every)
     wall = time.perf_counter() - t0
     print('ckpt=%s n=%d K=%d commit=%d fb=%.2f ki=%.2f select=%s replan=%s safety=%d' %
           (a.ckpt, r['n'], a.K, a.commit, a.fb, a.ki, a.select, a.replan, a.safety))

ml/sim/closed_loop.py

@@ -153,7 +153,7 @@ class TeacherOracle:
 def rollout(model, kin, tasks, device, K=8, safety=False, seed=0, max_sec=None, return_traj=False, use_sub=True, sub_override=None,
             qp0=None, steps=1, z_scale=1.0, select='consist', oracle=None, replan='r3', sub_rule='cross',
             fb_gain=0.0, fb_scale=0.0, fb_near=0.2, fb_fallback=0.0, safety_partial=0, commit_n=3, collect=None,
-            fb_exec=0.0, exec_safety=0, fb_ki=0.0, fb_decay=0.9, fb_imax=0.5):
+            fb_exec=0.0, exec_safety=0, fb_ki=0.0, fb_decay=0.9, fb_imax=0.5, fb_every=1):
     """use_sub=False：不用子目标（sf=0）；sub_override=(subs [B,M,6] 绝对关节角, n [B] 个数 或 need [B] bool)：冷启动时用给定途经点代替子目标头；
     qp0 [B,6]：起点的上一帧（默认 = q0，即静止起步）。steps：MeanFlow 采样段数；z_scale：噪声缩放；
     select：consist（部署口径：冷启动价值前瞻、之后与上一条已提交计划的一致性）/ value（每次都价值前瞻）/ first（不选，取第 0 条）。
@@ -173,6 +173,7 @@ def rollout(model, kin, tasks, device, K=8, safety=False, seed=0, max_sec=None,
     queue = torch.zeros(B, 0, 6, device=device)
     last = None; last_ok = torch.zeros(B, dtype=torch.bool, device=device)
     fb_i = torch.zeros(B, 6, device=device)   # 积分项：累积持续同向的目标反馈（慢变偏差）
+    fb_cache = None; pit = 0                  # fb_every>1：复用上一次规划的目标反馈（误差方向慢变）
     subs = torch.zeros(B, 2, 6, device=device); nsub = torch.zeros(B, dtype=torch.long, device=device)
     sub_t = torch.zeros(B, device=device); sub_prev = q0.clone()
     done = torch.zeros(B, dtype=torch.bool, device=device); t_done = torch.full((B,), float('nan'), device=device)
@@ -204,7 +205,11 @@ def rollout(model, kin, tasks, device, K=8, safety=False, seed=0, max_sec=None,
                 act, val, sub = model.sample_k(feat, K, z, steps=steps)
             inc = act.float().clamp(-1, 1) * astep
             if fb_gain > 0:   # 实验（2026-10-20）：目标反馈加在所有候选上（安全层也评估带反馈的候选）
-                fb = dls_steps(kin, qe, T, R, n=1)
+                if fb_cache is None or fb_every <= 1 or pit % fb_every == 0:
+                    fb = dls_steps(kin, qe, T, R, n=1); fb_cache = fb
+                else:
+                    fb = fb_cache
+                pit += 1
                 if fb_ki > 0:
                     fb_i = fb_decay * fb_i + fb[:, 0] * (~done)[:, None]
                     lim = step * fb_imax