refactor(alert): move AHDC track-finding from AHDCEngine to ALERTEngine

reconstruction/alert/src/main/java/org/jlab/rec/ahdc/Cluster/Cluster.java → reconstruction/alert/src/main/java/org/jlab/rec/ahdc/AHDCCluster/AHDCCluster.java

@@ -1,4 +1,4 @@
-package org.jlab.rec.ahdc.Cluster;
+package org.jlab.rec.ahdc.AHDCCluster;
 
 import org.jlab.rec.ahdc.Hit.Hit;
 import org.jlab.rec.ahdc.PreCluster.PreCluster;
@@ -9,9 +9,9 @@
 import org.jlab.geom.prim.Point3D;
 
 /**
- * Cluster are compose by 2 PreCluster on layer with a different stereo angle
+ * AHDCCluster are compose by 2 PreCluster on layer with a different stereo angle
  */
-public class Cluster {
+public class AHDCCluster {
 
 	private int                   _trackId = -1;
 	private double                _Radius;
@@ -50,7 +50,7 @@ private static double stereoTwistFromLine(Line3D line) {
 
 		return wrapPi(phi1 - phi0);
 	}
-	public Cluster(PreCluster precluster, PreCluster other_precluster) {
+	public AHDCCluster(PreCluster precluster, PreCluster other_precluster) {
 		this._PreClusters_list = new ArrayList<>();
 		_PreClusters_list.add(precluster);
 		_PreClusters_list.add(other_precluster);
@@ -80,15 +80,44 @@ public Cluster(PreCluster precluster, PreCluster other_precluster) {
 		this._V        = this._Y / (this._X * this._X + this._Y * this._Y);
 	}
 
-	public Cluster(double X, double Y, double Z) {
+	public AHDCCluster(double X, double Y, double Z) {
 		this._X = X;
 		this._Y = Y;
 		this._Z = Z;
 	}
 
+	/** Build an AHDCCluster from a single PreCluster (one layer of a superlayer).
+	 *  Used by the GNN path when a track covers a superlayer on only one
+	 *  stereo layer — no stereo pair is available, so Z is taken from the
+	 *  average wire-midpoint z of the PreCluster's hits rather than from a
+	 *  stereo-angle computation. DocaClusterRefiner falls back to a degenerate
+	 *  DocaCluster when {@code get_PreClusters_list().size() != 2}, so
+	 *  downstream is unaffected. */
+	public AHDCCluster(PreCluster precluster) {
+		this._PreClusters_list = new ArrayList<>();
+		_PreClusters_list.add(precluster);
+		this._Radius   = precluster.get_Radius();
+		this._Phi      = precluster.get_Phi();
+		this._X        = precluster.get_X();
+		this._Y        = precluster.get_Y();
+		this._Num_wire = (int) precluster.get_Num_wire();
+		double r2 = this._X * this._X + this._Y * this._Y;
+		if (r2 > 0.0) {
+			this._U = this._X / r2;
+			this._V = this._Y / r2;
+		}
+		double zSum = 0.0;
+		int    zCount = 0;
+		for (Hit h : precluster.get_hits_list()) {
+			Line3D line = h.getLine();
+			if (line != null) { zSum += line.midpoint().z(); zCount++; }
+		}
+		this._Z = (zCount > 0) ? zSum / zCount : 0.0;
+	}
+
 	@Override
 	public String toString() {
-		return "Cluster{" + "_X=" + _X + ", _Y=" + _Y + ", _Z=" + _Z + '}';
+		return "AHDCCluster{" + "_X=" + _X + ", _Y=" + _Y + ", _Z=" + _Z + '}';
 	}
 
 	public ArrayList<PreCluster> get_PreClusters_list() {

reconstruction/alert/src/main/java/org/jlab/rec/ahdc/Cluster/ClusterFinder.java → reconstruction/alert/src/main/java/org/jlab/rec/ahdc/AHDCCluster/AHDCClusterFinder.java

@@ -1,21 +1,21 @@
-package org.jlab.rec.ahdc.Cluster;
+package org.jlab.rec.ahdc.AHDCCluster;
 
 import org.jlab.rec.ahdc.PreCluster.PreCluster;
 
 import java.util.ArrayList;
 import java.util.List;
 
-/** ClusterFinder
+/** AHDCClusterFinder
  *
  * \todo description of what it does and how it works
  *
  */
-public class ClusterFinder {
+public class AHDCClusterFinder {
 
-	private final ArrayList<Cluster> _AHDCClusters                 = new ArrayList<>();
-	private final ArrayList<Cluster> _list_with_maybe_same_cluster = new ArrayList<>();
+	private final ArrayList<AHDCCluster> _AHDCClusters                 = new ArrayList<>();
+	private final ArrayList<AHDCCluster> _list_with_maybe_same_cluster = new ArrayList<>();
 
-	public ClusterFinder() {}
+	public AHDCClusterFinder() {}
 
 	private void find_associate_cluster(PreCluster precluster, List<PreCluster> AHDC_precluster_list, int window, int minimal_distance, int super_layer, int layer, int associate_super_layer) {
 	    //System.out.println(" precluster superlayer " + precluster.get_Super_layer() + " ref superlayer " + super_layer + " layer " + precluster.get_Layer() + " ref " + layer);
@@ -52,7 +52,7 @@ private void find_associate_cluster(PreCluster precluster, List<PreCluster> AHDC
 				if (best_precluster != null) {
 					precluster.set_Used(true);
 					best_precluster.set_Used(true);
-					Cluster new_Cluster = new Cluster(precluster, best_precluster);
+					AHDCCluster new_Cluster = new AHDCCluster(precluster, best_precluster);
 					_list_with_maybe_same_cluster.add(new_Cluster);
 				}
 			}
@@ -81,18 +81,18 @@ public void findCluster(List<PreCluster> AHDC_precluster_list) {
 			find_associate_cluster(precluster, AHDC_precluster_list, window, minimal_distance, 4, 2, 5);
 		}
 
-		for (Cluster cluster : _list_with_maybe_same_cluster) {
+		for (AHDCCluster cluster : _list_with_maybe_same_cluster) {
 			if (!containsCluster(_AHDCClusters, cluster.get_Phi(), cluster.get_Radius())) {
 				_AHDCClusters.add(cluster);
 			}
 		}
 	}
 
-	public boolean containsCluster(final List<Cluster> list, double phi, double radius) {
+	public boolean containsCluster(final List<AHDCCluster> list, double phi, double radius) {
 		return list.stream().anyMatch(o -> o.get_Radius() == (radius) && o.get_Phi() == phi);
 	}
 
-	public ArrayList<Cluster> get_AHDCClusters() {
+	public ArrayList<AHDCCluster> get_AHDCClusters() {
 		return _AHDCClusters;
 	}
 

reconstruction/alert/src/main/java/org/jlab/rec/ahdc/AI/GNNConstants.java

@@ -0,0 +1,42 @@
+package org.jlab.rec.ahdc.AI;
+
+/** Normalization and graph-construction constants for the GNN track finder.
+ *  Mirrors track-finding/gnn/config.py — keep in sync with the training config.
+ */
+final class GNNConstants {
+    private GNNConstants() {}
+
+    static final int   NODE_FEAT_DIM      = 10;
+    static final int   EDGE_FEAT_DIM      = 9;
+
+    // Model architecture parameters (control the minimum graph size at inference).
+    // GravNet progressive-k reaches 2*k, topk uses k+1 → N_nodes >= 2*k + 2.
+    // The exported model clamps topk(k+1) to N internally (see
+    // track-finding/export_torchscript.py::_knn_indices), so any graph with
+    // >=3 nodes runs without crashing. Smaller graphs can't form any edge
+    // with the MAX_LAYER_GAP rule anyway, so we skip them here.
+    static final int   MIN_NODES          = 3;
+
+    // Graph construction
+    static final int   MAX_LAYER_GAP      = 2;
+    static final double MAX_EDGE_DISTANCE = 35.0;              // mm
+    static final double MAX_EDGE_DIST_SQ  = MAX_EDGE_DISTANCE * MAX_EDGE_DISTANCE;
+
+    // Feature normalization
+    static final double MAX_R             = 100.0;             // mm
+    static final double DOCA_STD          = 10.0;              // mm
+    static final double Z_HALF_LENGTH     = 200.0;             // mm
+    static final double STEREO_ANGLE_MAX  = 0.03;              // rad
+    static final double STEREO_SCALE      = 1.0 / STEREO_ANGLE_MAX;
+
+    // ATOF abs_layer convention from Python's build_graph
+    static final int   ATOF_BAR_ABS_LAYER   = 10;   // component == 10
+    static final int   ATOF_WEDGE_ABS_LAYER = 11;   // all other components
+
+    // Track extraction: connected components at a single score threshold, matching
+    // gnn/evaluate.py (extract_tracks(..., method="cc", threshold=0.1)). Drop tracks
+    // with fewer than MIN_TRACK_NODES total nodes — same filter evaluate.py applies
+    // after the method call.
+    static final double TRACK_SCORE_THRESHOLD = 0.1;
+    static final int    MIN_TRACK_NODES       = 3;
+}