CLAUDE: leanMeasure - drop the just-added JCuda Java-oracle fallback branch...

CLAUDE: leanMeasure - drop the just-added JCuda Java-oracle fallback branch (Andrey's ruling: fallbacks only where free)

The dual-path else-branch added earlier today (f8df8dce) would silently run
a second implementation on the JCuda backend - the exact divergence the
JNA-migration-keep-it-clean policy avoids. Now: no GPU peak kernel -> loud
'requires the JNA backend' message, scene returns null. NOT a JCuda purge:
all pre-existing JCuda code stays (base execCorr2DPeaks null-return, the
consolidation CPU bridge, legacy paths); the double kernel and the Java
oracle remain as VALIDATION tiers in tests/export capture.

mvn package OK.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

src/main/java/com/elphel/imagej/cuas/rt/CuasPoseRT.java

@@ -445,12 +445,12 @@ public class CuasPoseRT {
 		// 6. peak + eigen per tile - the frozen 15-arg getMaxXYCmEig contract (Andrey's
 		// rulings 07/13-14/2026: closed-form 2x2 eigen; recentering + envelope de-bias as
 		// runtime params; FLOAT kernel in production - real-case float-vs-double dx/dy
-		// <= 7.2e-7 px, ratified 07/14). PRODUCTION PATH (JNA): corr2D_peak_eig_f measures
-		// on the GPU and only [n_corr][8] peak floats cross D2H - the PD download happens
-		// ONLY for debug rendering, the pose_corr capture, or the JCuda fallback (the
-		// original Java-oracle loop, results equal to the float kernel within validation
-		// tiers). By Claude on 07/14/2026 (was the unconditional PD D2H + Java loop).
-		final int corr_width = 2 * GPUTileProcessorDttSize() - 1;
+		// <= 7.2e-7 px, ratified 07/14). corr2D_peak_eig_f measures on the GPU and only
+		// [n_corr][8] peak floats cross D2H - the PD download happens ONLY for debug
+		// rendering or the pose_corr capture. NO JCuda fallback (Andrey's ruling 07/14:
+		// keep fallbacks only where free) - a backend without the kernel fails loudly;
+		// the double kernel + the Java oracle remain as VALIDATION tiers (tests/export).
+		// By Claude on 07/14/2026 (was the unconditional PD D2H + Java loop).
 		final double [] debias = (clt_parameters.curt.pose_debias > 0) ?
 				windowAutocorrDebias(GPUTileProcessorDttSize(), clt_parameters.curt.pose_debias) : null;
 		float [] fdebias = null;
@@ -470,7 +470,12 @@ public class CuasPoseRT {
 				clt_parameters.imp.eig_scale_axes,// recenter window: scale eigen-ellipse half-axes
 				clt_parameters.imp.eig_inc_axes,  // recenter window: add to half-axes, pix
 				true);                            // float = production (double stays the validation tier)
-		if ((gpu_peaks != null) && !peaks_path_reported) {
+		if (gpu_peaks == null) { // no fallback by ruling - fail loudly, not silently slower
+			System.out.println("leanMeasure(): GPU peak kernel unavailable (JCuda backend?) - "+
+					"the lean pose chain requires the JNA backend (-Dtp.backend=jna)");
+			return null;
+		}
+		if (!peaks_path_reported) {
 			peaks_path_reported = true;
 			System.out.println("CuasPoseRT: GPU peak kernel active (corr2D_peak_eig_f, float; "+
 					"recenter="+clt_parameters.curt.pose_recenter+", debias="+clt_parameters.curt.pose_debias+
@@ -478,7 +483,7 @@ public class CuasPoseRT {
 		}
 		// PD crosses only when actually consumed
 		double [][] corr_pd = null;
-		if ((gpu_peaks == null) || (corr_pd_out != null) || PoseCorrExport.isArmed()) {
+		if ((corr_pd_out != null) || PoseCorrExport.isArmed()) {
 			corr_pd = TDCorrTile.fetchNormalizedPD(gpuQuad, corr_indices);
 			if (corr_pd == null) {
 				System.out.println("leanMeasure(): fetchNormalizedPD returned null for scene "+scene.getImageName());
@@ -496,57 +501,20 @@ public class CuasPoseRT {
 		final double [][] vector_XYS = new double [num_pix][];
 		final double [][] eigen =      new double [num_pix][];
 		int num_meas = 0;
-		if (gpu_peaks != null) {
-			// production path: peak rows align 1:1 with corr_indices; sum slots carry the
-			// measurement, valid==0 = kernel reject (weak/NaN/degenerate, as the oracle null)
-			for (int n = 0; n < corr_indices.length; n++) {
-				if ((corr_indices[n] & ((1 << GPUTileProcessor.CORR_NTILE_SHIFT) - 1)) != 0xff) continue;
-				final int i = corr_indices[n] >> GPUTileProcessor.CORR_NTILE_SHIFT;
-				if ((i < 0) || (i >= num_pix) || (pXpYD_center[i] == null)) continue;
-				final int row = n * GpuQuad.PEAK_ROW_FLOATS;
-				if (gpu_peaks[row + GpuQuad.PEAK_ROW_FLOATS - 1] == 0.0f) continue; // rejected
-				centers[i] =    pXpYD_center[i];
-				vector_XYS[i] = new double [] {gpu_peaks[row], gpu_peaks[row + 1], gpu_peaks[row + 2]};
-				eigen[i] =      new double [] {gpu_peaks[row + 3], gpu_peaks[row + 4],
-						gpu_peaks[row + 5], gpu_peaks[row + 6]};
-				num_meas++;
-			}
-		} else {
-			// JCuda fallback: the original Java-oracle loop (identical contract, double math)
-			for (int i = 0; i < num_pix; i++) {
-				if ((i >= corr_pd.length) || (corr_pd[i] == null) || (pXpYD_center[i] == null)) continue;
-				double [] corr_tile = corr_pd[i]; // not modified by getMaxXYCmEig (fpn_mask == null)
-				if (debias != null) { // per-tile copy: corr_pd stays raw for -POSE-RT-CORR2D // By Claude on 07/13/2026
-					corr_tile = new double [corr_pd[i].length];
-					for (int k = 0; k < corr_tile.length; k++) {
-						corr_tile[k] = corr_pd[i][k] * debias[k];
-					}
-				}
-				final double [] r = Correlation2d.getMaxXYCmEig(
-						corr_tile,
-						corr_width,
-						clt_parameters.imp.eig_min_abs,   // abs_min
-						clt_parameters.imp.eig_min_rel,   // rel_min
-						clt_parameters.imp.min_str_sum,   // min_peak (single channel = the "sum")
-						clt_parameters.imp.eig_sub_frac,  // pedestal suppression - was hardcoded 0.0: pedestal-inflated
-						                                  // eigen lambdas -> sqrt(l) hits eig_max_sqrt -> setEigenTransform
-						                                  // k=max(0,1/sqrt(l)-1/max) ZEROES wide-peak tiles (the eigen-ON
-						                                  // 3x slow convergence). By Claude on 07/12/2026, Andrey's go.
-						clt_parameters.curt.pose_recenter,// recentering passes (0 = baseline single-pass) // By Claude on 07/13/2026
-						clt_parameters.imp.eig_sub_frac1, // pedestal suppression during recentering
-						clt_parameters.imp.eig_scale_axes,// recenter window: scale eigen-ellipse half-axes
-						clt_parameters.imp.eig_inc_axes,  // recenter window: add to half-axes, pix
-						null,                             // fpn_mask (input is FPN-subtracted)
-						false,                            // ignore_border
-						null,                             // debug_data
-						true,                             // eig_fast2x2 FROZEN true: the closed-form eigen IS the contract
-						false);                           // debug
-				if (r == null) continue;
-				centers[i] =    pXpYD_center[i];
-				vector_XYS[i] = new double [] {r[0], r[1], r[2]};
-				eigen[i] =      new double [] {r[3], r[4], r[5], r[6]};
-				num_meas++;
-			}
+		// peak rows align 1:1 with corr_indices; sum slots carry the measurement,
+		// valid==0 = kernel reject (weak/NaN/degenerate, as the oracle null). The float
+		// values are widened to double ONCE here (exact) - the LMA solver stays double.
+		for (int n = 0; n < corr_indices.length; n++) {
+			if ((corr_indices[n] & ((1 << GPUTileProcessor.CORR_NTILE_SHIFT) - 1)) != 0xff) continue;
+			final int i = corr_indices[n] >> GPUTileProcessor.CORR_NTILE_SHIFT;
+			if ((i < 0) || (i >= num_pix) || (pXpYD_center[i] == null)) continue;
+			final int row = n * GpuQuad.PEAK_ROW_FLOATS;
+			if (gpu_peaks[row + GpuQuad.PEAK_ROW_FLOATS - 1] == 0.0f) continue; // rejected
+			centers[i] =    pXpYD_center[i];
+			vector_XYS[i] = new double [] {gpu_peaks[row], gpu_peaks[row + 1], gpu_peaks[row + 2]};
+			eigen[i] =      new double [] {gpu_peaks[row + 3], gpu_peaks[row + 4],
+					gpu_peaks[row + 5], gpu_peaks[row + 6]};
+			num_meas++;
 		}
 		if (debugLevel > -2) {
 			System.out.println("leanMeasure(): scene "+scene.getImageName()+" measured "+num_meas+" tiles");