feat(planning): consolidate recorded-route localization and spatial scene

Preserve the completed teach-and-repeat laboratory stage: reference preparation, cascaded acquisition, local tracking and recovery, recording lifecycle, replay qualification, and persistent Rerun scene controls. Document the open grid-picking regression and Rerun upgrade contract. No autonomous driving or loop-closure optimization is claimed.

src/k1link/missions/causal_tracking.py

@@ -0,0 +1,76 @@
+"""Experimental temporal qualification; never grants vehicle authority."""
+
+import numpy as np
+
+from .registration import angle_deg, rigid, transform
+
+TRACKING_POLICY = dict(
+    version="causal-consistency/v1",
+    consecutive=3,
+    maximum_position_change_m=0.5,
+    maximum_rotation_change_deg=5.0,
+    maximum_age_s=8.0,
+)
+
+
+class CausalTracking:
+    def __init__(self):
+        self.matrix = None
+        self.sample_ns = 0
+        self.segment = 0
+        self.streak = 0
+        self.state = "acquiring"
+        self.reason = "initial"
+
+    def clear(self, reason):
+        self.matrix = None
+        self.sample_ns = 0
+        self.streak = 0
+        self.state = "lost"
+        self.reason = reason
+
+    def tick(self, now_ns, segment):
+        if segment != self.segment:
+            self.clear("receipt-gap")
+            self.segment = segment
+        elif (
+            self.matrix is not None
+            and (now_ns - self.sample_ns) / 1e9 > TRACKING_POLICY["maximum_age_s"]
+        ):
+            self.clear("stale")
+
+    def accept(self, result, sample, now_ns, segment):
+        self.tick(now_ns, segment)
+        age = (now_ns - sample["monotonic_ns"]) / 1e9
+        evidence = dict(age_s=age, position_change_m=None, rotation_change_deg=None)
+        if sample["segment"] != segment:
+            # An old job must never overwrite new-segment state.
+            return {**evidence, "accepted": False, "reason": "old-segment"}
+        if not 0 <= age <= TRACKING_POLICY["maximum_age_s"]:
+            self.clear("stale-result")
+        elif result["status"] != "candidate":
+            self.clear("registration-rejected")
+        else:
+            matrix = rigid(result["T_reference_query"])
+            if self.matrix is not None:
+                position = sample["path"][-1:]
+                delta = float(
+                    np.linalg.norm(transform(position, matrix) - transform(position, self.matrix))
+                )
+                rotation = angle_deg(matrix[:3, :3] @ self.matrix[:3, :3].T)
+                evidence.update(position_change_m=delta, rotation_change_deg=rotation)
+                if (
+                    delta > TRACKING_POLICY["maximum_position_change_m"]
+                    or rotation > TRACKING_POLICY["maximum_rotation_change_deg"]
+                ):
+                    self.clear("inconsistent-candidate")
+                    return {**evidence, "accepted": False, "reason": self.reason}
+            self.matrix = matrix
+            self.sample_ns = sample["monotonic_ns"]
+            self.streak += 1
+            self.state = (
+                "tracking" if self.streak >= TRACKING_POLICY["consecutive"] else "acquiring"
+            )
+            self.reason = "consistent-candidate"
+            return {**evidence, "accepted": True, "reason": self.reason}
+        return {**evidence, "accepted": False, "reason": self.reason}