feat(plan): recursive plan revision — living plan document (N17)

When readiness≥0.7 and conviction changes significantly mid-trade:
+30% conviction → extend plan (thesis strengthening)
-60% conviction → shorten to exit in 3 bars (thesis collapsing)
Normal fluctuation (0.4x-1.3x) → plan unchanged.
The model learns to revise plans based on evolving assessment.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

crates/ml/src/cuda_pipeline/experience_kernels.cu

@@ -1485,11 +1485,31 @@ extern "C" __global__ void experience_env_step(
         position *= fmaxf(ps[27] * readiness, 0.1f);
     }
 
+    int has_plan = (ps[23] > 0.5f);
+
+    /* Recursive plan revision (N17): living plan document.
+     * When model is highly confident (readiness ≥ 0.7) and conviction changes
+     * significantly, update the stored plan. The model learns WHEN to revise. */
+    if (has_plan && fabsf(position) > 0.001f && readiness >= 0.7f && plan_params_ptr != NULL) {
+        float conv_now = plan_params_ptr[i * 6 + 4];
+        float conv_entry = ps[27];
+
+        if (conv_now > conv_entry * 1.3f) {
+            /* Conviction increased 30%+ → thesis strengthening → extend plan */
+            float new_target = plan_params_ptr[i * 6 + 0];
+            ps[23] = fmaxf(ps[23], new_target);  /* only extend, never shorten from conviction increase */
+            ps[24] = fmaxf(ps[24], plan_params_ptr[i * 6 + 1]);  /* widen profit target */
+        } else if (conv_now < conv_entry * 0.4f) {
+            /* Conviction dropped 60%+ → thesis collapsing → shorten plan to exit soon */
+            ps[23] = fminf(ps[23], hold_time + 3.0f);  /* exit within 3 bars */
+        }
+        /* Between 0.4x and 1.3x → plan stays as-is (normal fluctuation) */
+    }
+
     /* Plan enforcement: ISV-driven safety net.
      * The model LEARNS exit timing through Q-values + conviction-scaled reward.
      * Safety thresholds are ISV-modulated — tighter when unstable, wider when stable.
      * All driven by temporal attention, nothing hardcoded. */
-    int has_plan = (ps[23] > 0.5f);
     if (has_plan && fabsf(position) > 0.001f) {
         float pnl_pct = (raw_close - entry_price) / fmaxf(fabsf(entry_price), 1.0f)
                          * ((position > 0.0f) ? 1.0f : -1.0f);