Add distributed server implementation and protocol definitions

- Introduced `protocol.py` for shared data models used in server/client communication, including request and response schemas for registration, job submission, and status updates.
- Implemented `server.py` to manage a SQLite job queue and client registry, handling job polling, status updates, and job completion.
- Created a cheat sheet for server usage, detailing commands for starting the server, submitting jobs, and monitoring clients.
- Added several experiment configuration files for various training setups, including geometry vector injections and baseline ensembles.

.gitignore

@@ -15,3 +15,6 @@ scripts/deprecated/
 v3/results/*
 scripts/utility/backup_mirror_with_archive.sh
 v3/distributed/logs/*
+v4/configs/**/
+v4/distributed/logs/*
+v4/results/*

v3/classes/geometry_towers.py

@@ -17,6 +17,7 @@ SegMapDataset          — Dataset yielding (seg_tensor, label) pairs
 SegCNN                 — pretrained CNN adapted for segmentation-map input
 GeometryTower          — TowerBase implementation (the main class to use)
 """
+
 from __future__ import annotations
 
 from dataclasses import dataclass
@@ -35,15 +36,18 @@ from tqdm import tqdm
 import pandas as pd
 
 from v3.classes.towerbase import TowerBase
+from v3.classes.towerbase import TowerBase, EarlyPassContext
 
 
 # ---------------------------------------------------------------------------
 # Data record
 # ---------------------------------------------------------------------------
 
+
 @dataclass
 class SegMapRecord:
     """One labelled eye sample for the seg-map CNN."""
+
     sample_id: str
     image_path: Path  # original fundus image (used by unet mode)
     annotation_disc: Path  # contour (.txt) or mask (.bmp/.png)
@@ -59,6 +63,7 @@ class SegMapRecord:
 # Seg-map utilities
 # ---------------------------------------------------------------------------
 
+
 def _combine_masks(disc_mask: np.ndarray, cup_mask: np.ndarray) -> np.ndarray:
     """Combine binary disc and cup masks into a 3-class label map.
 
@@ -69,7 +74,7 @@ def _combine_masks(disc_mask: np.ndarray, cup_mask: np.ndarray) -> np.ndarray:
     """
     disc = (disc_mask > 0).astype(np.uint8)
     cup = (cup_mask > 0).astype(np.uint8)
-    cup  = (cup & disc)          # structural prior: cup must be inside disc
+    cup = cup & disc  # structural prior: cup must be inside disc
     seg = disc + cup  # 0, 1 (rim), or 2 (cup)
     return seg.astype(np.uint8)
 
@@ -86,7 +91,7 @@ def crop_to_disc(seg_map: np.ndarray) -> np.ndarray:
         return seg_map
     r0, r1 = int(np.argmax(rows)), int(len(rows) - 1 - np.argmax(rows[::-1]))
     c0, c1 = int(np.argmax(cols)), int(len(cols) - 1 - np.argmax(cols[::-1]))
-    return seg_map[r0:r1 + 1, c0:c1 + 1]
+    return seg_map[r0 : r1 + 1, c0 : c1 + 1]
 
 
 def seg_map_to_tensor(
@@ -120,12 +125,15 @@ def seg_map_to_tensor(
 # GT mask loading (pure NumPy / PIL — no CUDA, safe in DataLoader workers)
 # ---------------------------------------------------------------------------
 
+
 def _load_contour(path: Path) -> np.ndarray:
     """Load x,y contour pairs from a whitespace- or comma-delimited text file."""
     arr = np.zeros((0, 2), dtype=np.float32)
     for delimiter in (",", None):
         try:
-            candidate = np.loadtxt(str(path), delimiter=delimiter, comments="#", dtype=np.float32)
+            candidate = np.loadtxt(
+                str(path), delimiter=delimiter, comments="#", dtype=np.float32
+            )
             if candidate.size > 0:
                 arr = candidate
                 break
@@ -169,7 +177,9 @@ def _extract_masks_from_image(
 
     if arr.ndim == 2:
         edges = np.concatenate([arr[0], arr[-1], arr[:, 0], arr[:, -1]])
-        bg_val = int(np.argmax(np.bincount(edges.astype(np.int64).clip(0, 255), minlength=256)))
+        bg_val = int(
+            np.argmax(np.bincount(edges.astype(np.int64).clip(0, 255), minlength=256))
+        )
         disc_arr = (arr != bg_val).astype(np.uint8)
         vals = np.unique(arr)
         non_bg = vals[vals != bg_val]
@@ -184,7 +194,9 @@ def _extract_masks_from_image(
         arr = np.array(img_rgb)
         h, w, c = arr.shape
         edges_rgb = np.concatenate([arr[0], arr[-1], arr[:, 0], arr[:, -1]], axis=0)
-        edge_colors, edge_counts = np.unique(edges_rgb.reshape(-1, c), axis=0, return_counts=True)
+        edge_colors, edge_counts = np.unique(
+            edges_rgb.reshape(-1, c), axis=0, return_counts=True
+        )
         bg_color = edge_colors[int(np.argmax(edge_counts))]
         colors, counts = np.unique(arr.reshape(-1, c), axis=0, return_counts=True)
         not_bg = np.any(colors != bg_color.reshape(1, -1), axis=1)
@@ -221,7 +233,9 @@ def load_gt_masks(rec: SegMapRecord, target_size: int) -> Tuple[np.ndarray, np.n
 
     # ---- Disc ----
     if rec.annotation_type_disc == "mask":
-        disc_mask, cup_from_disc = _extract_masks_from_image(rec.annotation_disc, target_size)
+        disc_mask, cup_from_disc = _extract_masks_from_image(
+            rec.annotation_disc, target_size
+        )
         if cup_from_disc.any():
             cup_mask = cup_from_disc
     else:  # contour
@@ -250,6 +264,7 @@ def load_gt_masks(rec: SegMapRecord, target_size: int) -> Tuple[np.ndarray, np.n
 # U-Net fine-tuning dataset
 # ---------------------------------------------------------------------------
 
+
 class UNetFineTuneDataset(Dataset):
     """Loads (image_tensor, mask_tensor) pairs for fine-tuning the U-Net."""
 
@@ -294,6 +309,7 @@ class UNetFineTuneDataset(Dataset):
 # U-Net precomputation
 # ---------------------------------------------------------------------------
 
+
 def precompute_unet_seg_maps(
     records: List[SegMapRecord],
     segmenter,
@@ -313,7 +329,7 @@ def precompute_unet_seg_maps(
         probs = torch.sigmoid(logits)[0].cpu().numpy()
         disc = (probs[0] > threshold).astype(np.uint8)
         cup = (probs[1] > threshold).astype(np.uint8)
-        cup  = (cup & disc)
+        cup = cup & disc
         seg_maps.append(_combine_masks(disc, cup.astype(np.uint8)))
     return seg_maps
 
@@ -322,6 +338,7 @@ def precompute_unet_seg_maps(
 # SegMapDataset
 # ---------------------------------------------------------------------------
 
+
 class SegMapDataset(Dataset):
     """PyTorch Dataset that yields (seg_tensor, label) pairs."""
 
@@ -445,7 +462,9 @@ class SegCNN(nn.Module):
             if pretrained:
                 with torch.no_grad():
                     new_conv.weight.copy_(
-                        first_conv.weight.mean(dim=1, keepdim=True).expand_as(new_conv.weight)
+                        first_conv.weight.mean(dim=1, keepdim=True).expand_as(
+                            new_conv.weight
+                        )
                     )
             self._replace_first_conv(base, new_conv)
 
@@ -486,6 +505,7 @@ class SegCNN(nn.Module):
 # GeometryTower — TowerBase implementation
 # ---------------------------------------------------------------------------
 
+
 class GeometryTower(TowerBase, nn.Module):
     """TowerBase implementation for the optic-disc/cup segmentation modality.
 
@@ -578,6 +598,48 @@ class GeometryTower(TowerBase, nn.Module):
         for p in self._seg_cnn.parameters():
             p.requires_grad = trainable
 
+    def early_pass(self, context: EarlyPassContext) -> None:
+        """
+        V4 orchestrator hook: combines prepare_fold and augment_samples.
+
+        1. Pre-computes all segmentation maps for the current fold.
+        2. Injects them into the sample dicts held by the context object.
+        """
+        if self._manifest_path is None:
+            raise ValueError("GeometryTower requires manifest_path")
+
+        all_paths: dict = {}
+        for split in (
+            context.eye_train,
+            context.bilat_train,
+            context.bilat_val,
+            context.bilat_test,
+        ):
+            for s in split:
+                for slot in ("image_1", "image_2"):
+                    p = s.get(slot)
+                    if p is not None:
+                        all_paths[str(Path(p).resolve())] = None
+
+        # 1. Pre-compute seg maps (from prepare_fold)
+        if self._geometry_source == "unet":
+            self._prepare_fold_unet(
+                list(all_paths.keys()), context.eye_train, context.device
+            )
+        else:
+            self._prepare_fold_gt(list(all_paths.keys()))
+
+        # 2. Inject into samples (from augment_samples)
+        for samples in (
+            context.eye_train,
+            context.bilat_train,
+            context.bilat_val,
+            context.bilat_test,
+        ):
+            if samples:
+                # This modifies the list of dicts in the context object in-place
+                self.augment_samples(samples)
+
     def prepare_fold(
         self,
         *,
@@ -617,7 +679,10 @@ class GeometryTower(TowerBase, nn.Module):
             (self._in_channels, self._target_size, self._target_size), dtype=np.float32
         )
         for s in samples:
-            for img_slot, seg_slot in (("image_1", "seg_map_1"), ("image_2", "seg_map_2")):
+            for img_slot, seg_slot in (
+                ("image_1", "seg_map_1"),
+                ("image_2", "seg_map_2"),
+            ):
                 img_path = s.get(img_slot)
                 if img_path is None:
                     continue
@@ -687,7 +752,10 @@ class GeometryTower(TowerBase, nn.Module):
                 n_ok += 1
             except Exception:
                 self._seg_cache[img_path] = self._seg_map_to_array(blank)
-        print(f"[GeometryTower] {n_ok}/{len(image_paths)} GT seg maps computed", flush=True)
+        print(
+            f"[GeometryTower] {n_ok}/{len(image_paths)} GT seg maps computed",
+            flush=True,
+        )
 
     def _prepare_fold_unet(self, image_paths: list, eye_train: list, device) -> None:
         """Pre-compute U-Net seg maps, with optional per-fold fine-tuning."""
@@ -716,14 +784,20 @@ class GeometryTower(TowerBase, nn.Module):
                     target_size=segmenter.target_size,
                     normalize=self._unet_normalize,
                 ),
-                batch_size=4, shuffle=True, num_workers=0,
+                batch_size=4,
+                shuffle=True,
+                num_workers=0,
+            )
+            optimizer = torch.optim.Adam(
+                segmenter.model.parameters(), lr=self._finetune_unet_lr
             )
-            optimizer = torch.optim.Adam(segmenter.model.parameters(), lr=self._finetune_unet_lr)
             criterion = torch.nn.BCEWithLogitsLoss()
             segmenter.model.train()
             for _ in range(self._finetune_unet_epochs):
                 for images, masks in ft_loader:
-                    images, masks = images.to(segmenter.device), masks.to(segmenter.device)
+                    images, masks = images.to(segmenter.device), masks.to(
+                        segmenter.device
+                    )
                     optimizer.zero_grad()
                     criterion(segmenter.model(images), masks).backward()
                     optimizer.step()
@@ -735,10 +809,15 @@ class GeometryTower(TowerBase, nn.Module):
         )
         records = [
             SegMapRecord(
-                sample_id="", image_path=Path(p),
-                annotation_disc=Path(p), annotation_cup=Path(p),
-                annotation_type_disc="", annotation_type_cup="",
-                patient_id=0, eye="", label=0,
+                sample_id="",
+                image_path=Path(p),
+                annotation_disc=Path(p),
+                annotation_cup=Path(p),
+                annotation_type_disc="",
+                annotation_type_cup="",
+                patient_id=0,
+                eye="",
+                label=0,
             )
             for p in image_paths
         ]
@@ -764,7 +843,8 @@ class GeometryTower(TowerBase, nn.Module):
                 entry = manifest_index.get(key)
                 if entry is None:
                     continue
-                records.append(SegMapRecord(
+                records.append(
+                    SegMapRecord(
                         sample_id="",
                         image_path=Path(p),
                         annotation_disc=Path(entry["annotation_disc"]),
@@ -774,5 +854,6 @@ class GeometryTower(TowerBase, nn.Module):
                         patient_id=int(s.get("patient_id", 0)),
                         eye=str(s.get("eye", "")),
                         label=int(s.get("label", 0)),
-                ))
+                    )
+                )
         return records

v3/classes/results.py

@@ -98,6 +98,10 @@ class FoldResult:
     fused_val_n:         int          = 0
     fused_test_auc:      float        = float("nan")
     fused_test_acc:      float        = float("nan")
+    fused_test_kappa:    float        = float("nan")
+    fused_test_f1:       float        = float("nan")
+    fused_test_ece:      float        = float("nan")
+    fused_test_n:        int          = 0
 
 
 @dataclass
@@ -136,3 +140,5 @@ class FoldArtifacts:
     probs_test:          Optional[np.ndarray] = None
     probs_test_img:      Optional[np.ndarray] = None
     probs_test_md:       Optional[np.ndarray] = None
+    y_true_fused_test:   Optional[np.ndarray] = None
+    probs_fused_test:    Optional[np.ndarray] = None

v3/classes/v3_hypertower.py

@@ -467,6 +467,10 @@ class V3HyperTower:
                 np.save(fold_dir / "test_probs_img.npy", artifacts.probs_test_img)
             if artifacts.probs_test_md is not None:
                 np.save(fold_dir / "test_probs_cd.npy", artifacts.probs_test_md)
+            if artifacts.y_true_fused_test is not None:
+                np.save(fold_dir / "test_y_true_fused.npy", artifacts.y_true_fused_test)
+            if artifacts.probs_fused_test is not None:
+                np.save(fold_dir / "test_probs_fused_head.npy", artifacts.probs_fused_test)
 
         if pred_store is not None:
             pred_store.save(tm_dir / "predictions.npz")
@@ -489,14 +493,17 @@ class V3HyperTower:
         if tower_mode in ("single", "classic"):
             _test_key = "classic_test"
         elif tower_mode == "ensemble":
-            _test_key = "ensemble_test"
+            _test_key = "fused_test" if fused_head else "ensemble_test"
         elif tower_mode in ("bilateral", "siamese"):
             _test_key = "bilat_test"
         else:
             _test_key = "classic_test"
+        mode_summary = {_test_key: summary.get(_test_key, {})}
+        if fused_head and tower_mode == "ensemble":
+            mode_summary["ensemble_test"] = summary.get("ensemble_test", {})
+            mode_summary["fused_best_val"] = summary.get("fused_best_val", {})
         with (tm_dir / "summary.json").open("w") as f:
-            json.dump({"mode_summary": {_test_key: summary.get(_test_key, {})}},
-                      f, indent=2, default=str)
+            json.dump({"mode_summary": mode_summary}, f, indent=2, default=str)
 
         return out_dir
 
@@ -1524,10 +1531,17 @@ class V3HyperTower:
         y_fu_best = p_fu_best = None
         if run_fused and single is not None:
             y_fu_best, p_fu_best = collect_probs_fused(fused, val_loader, device)
+            if y_fu_best is not None and y_fu_best.size:
+                fu_acc_best = float((p_fu_best.argmax(1) == y_fu_best).mean())
+                snap_fused, _, _, _ = _tune_and_snap(
+                    y_fu_best, p_fu_best, fu_acc_best, num_classes, args, args.ece_bins
+                )
 
         # Test set evaluation (once, never seen during training)
         snap_test: dict = {}
+        snap_fused_test: dict = {}
         y_test_out = p_test_out = p_test_img_out = p_test_md_out = None
+        y_fused_test_out = p_fused_test_out = None
 
         if test_loader is not None:
             if run_single and tower_mode == "ensemble":
@@ -1564,6 +1578,33 @@ class V3HyperTower:
                     {"fused": p_test_out, "img": p_test_img_out, "md": p_test_md_out},
                     suffix="_test",
                 )
+            if run_fused and single is not None:
+                y_fused_test_out, p_fused_test_out = collect_probs_fused(
+                    fused, test_loader, device
+                )
+                if y_fused_test_out is not None and y_fused_test_out.size:
+                    fused_test_acc_raw = float(
+                        (p_fused_test_out.argmax(1) == y_fused_test_out).mean()
+                    )
+                    snap_fused_test, _, _, _ = _tune_and_snap(
+                        y_fused_test_out, p_fused_test_out, fused_test_acc_raw,
+                        num_classes, args, args.ece_bins
+                    )
+                    print(
+                        f"  [fold {fold+1}] FUSED_HEAD TEST  "
+                        f"auc={snap_fused_test.get('auc', nan):.2f}  "
+                        f"acc={snap_fused_test.get('acc', nan):.2f}  "
+                        f"kappa={snap_fused_test.get('kappa', nan):.2f}  "
+                        f"f1={snap_fused_test.get('macro_f1', nan):.2f}  "
+                        f"ece={snap_fused_test.get('ece', nan):.2f}  "
+                        f"n={snap_fused_test.get('n', 0)}",
+                        flush=True,
+                    )
+                    _save_predictions_csv(
+                        fold_dir, mode, y_fused_test_out,
+                        {"fused_head": p_fused_test_out},
+                        suffix="_fused_head_test",
+                    )
         else:
             print(f"  [fold {fold+1}] WARNING: no test samples for this fold.", flush=True)
 
@@ -1630,7 +1671,12 @@ class V3HyperTower:
             fused_val_threshold=snap_fused.get("threshold", nan),
             fused_val_bias=_svf(snap_fused.get("bias")),
             fused_val_n=snap_fused.get("n", 0),
-            fused_test_auc=nan, fused_test_acc=nan,
+            fused_test_auc=snap_fused_test.get("auc", nan),
+            fused_test_acc=snap_fused_test.get("acc", nan),
+            fused_test_kappa=snap_fused_test.get("kappa", nan),
+            fused_test_f1=snap_fused_test.get("macro_f1", nan),
+            fused_test_ece=snap_fused_test.get("ece", nan),
+            fused_test_n=snap_fused_test.get("n", 0),
         ), FoldArtifacts(
             y_true_classic=y_cl_best,   probs_classic=p_cl_best,
             y_true_ensemble=y_en_best,  probs_ensemble=p_en_best,
@@ -1657,6 +1703,8 @@ class V3HyperTower:
             probs_test=p_test_out,
             probs_test_img=p_test_img_out,
             probs_test_md=p_test_md_out,
+            y_true_fused_test=y_fused_test_out,
+            probs_fused_test=p_fused_test_out,
         )
 
     @staticmethod
@@ -1685,6 +1733,7 @@ class V3HyperTower:
             ("ensemble_test", "ensemble_test"),
             ("classic_test",  "classic_test"),
             ("bilat_test",    "bilat_test"),
+            ("fused_test",    "fused_test"),
         ]:
             sub = {}
             for m in ["auc", "acc", "kappa", "f1", "ece"]:

v4/classes/accessory/backbones.py

@@ -109,6 +109,12 @@ BACKBONES: Dict[str, BackboneSpec] = {
         strip=_strip_efficientnet,
         blocks=_blocks_efficientnet,
     ),
+    "resnet18": BackboneSpec(
+        ctor=models.resnet18,
+        weights_default=models.ResNet18_Weights.DEFAULT,
+        strip=_strip_resnet,
+        blocks=_blocks_resnet,
+    ),
     "resnet50": BackboneSpec(
         ctor=models.resnet50,
         weights_default=models.ResNet50_Weights.DEFAULT,

v4/classes/accessory/transforms.py

@@ -44,6 +44,37 @@ class ImageTransformConfig:
         ]
         return transforms.Compose(ops)
 
+    def build_precache(self) -> transforms.Compose:
+        """Deterministic prefix: PIL → resized CHW float32 in [0, 1].
+
+        Output is suitable for caching; per-batch ``build_postcache`` finishes
+        the pipeline (augment + normalize) on tensors.
+        """
+        return transforms.Compose([
+            transforms.Resize(self.resize_size),
+            transforms.CenterCrop(self.crop_size),
+            transforms.ToTensor(),
+        ])
+
+    def build_postcache(self) -> transforms.Compose:
+        """Per-batch tail run on cached float32 [0, 1] CHW tensors.
+
+        Augmentations operate on tensors (torchvision v1 supports this for
+        Flip/Rotation/ColorJitter on tensor input).  Normalize is applied last.
+        """
+        ops = []
+        if self.augment:
+            if self.hflip:
+                ops.append(transforms.RandomHorizontalFlip())
+            if self.vflip:
+                ops.append(transforms.RandomVerticalFlip())
+            if self.rotation_deg:
+                ops.append(transforms.RandomRotation(self.rotation_deg))
+            if self.color_jitter:
+                ops.append(transforms.ColorJitter(*self.color_jitter))
+        ops.append(transforms.Normalize(mean=self.mean, std=self.std))
+        return transforms.Compose(ops)
+
 
 def backbone_transform_config(backbone_name: str, augment: bool = True) -> ImageTransformConfig:
     """Build an ImageTransformConfig using the backbone's default normalisation stats."""
@@ -64,3 +95,15 @@ def build_backbone_transform(backbone_name: str, augment: bool = True) -> transf
 def build_eval_transform(backbone_name: str) -> transforms.Compose:
     """Deterministic eval transform — no augmentation, backbone-matched normalisation."""
     return build_backbone_transform(backbone_name, augment=False)
+
+
+def build_split_transforms(
+    backbone_name: str, augment: bool = True
+) -> tuple[transforms.Compose, transforms.Compose]:
+    """Return (precache, postcache) transform pair for tensor-cached image towers.
+
+    precache  : PIL → CHW float32 in [0, 1]   (deterministic, run once at fill)
+    postcache : tensor → augmented + normalized tensor   (run per batch)
+    """
+    cfg = backbone_transform_config(backbone_name, augment=augment)
+    return cfg.build_precache(), cfg.build_postcache()

v4/classes/accessory/unet.py

@@ -0,0 +1,212 @@
+"""unet — REFUGE-trained UNet wrapper for v4.
+
+Lean accessory module: model definition + a thin segmenter wrapper that handles
+weight loading, preprocessing, inference, and fine-tuning.
+
+Used by:
+  - GeometrySegEncoder tower (produces disc/cup seg maps as CNN input)
+  - (future) ImageEncoder cropping (locates disc bbox for image cropping)
+
+The segmenter is intentionally domain-agnostic: it takes PIL images in and
+returns binary (disc, cup) numpy masks.  Fine-tuning consumes any DataLoader
+yielding (image_tensor, mask_tensor) pairs — mask preparation (parsing GT
+contour files, etc.) lives in the consumer.
+"""
+from __future__ import annotations
+
+from pathlib import Path
+
+import numpy as np
+import torch
+from PIL import Image
+from PIL.Image import Resampling
+from torch import nn
+from torchvision import transforms
+
+
+# ---------------------------------------------------------------------------
+# Model
+# ---------------------------------------------------------------------------
+
+class UNet(nn.Module):
+    def __init__(self, in_channels: int = 3, base_channels: int = 32, out_channels: int = 2):
+        super().__init__()
+        self.enc1 = self._block(in_channels, base_channels)
+        self.enc2 = self._block(base_channels, base_channels * 2)
+        self.enc3 = self._block(base_channels * 2, base_channels * 4)
+        self.enc4 = self._block(base_channels * 4, base_channels * 8)
+
+        self.pool = nn.MaxPool2d(2)
+        self.bottleneck = self._block(base_channels * 8, base_channels * 16)
+
+        self.up4  = nn.ConvTranspose2d(base_channels * 16, base_channels * 8, 2, stride=2)
+        self.dec4 = self._block(base_channels * 16, base_channels * 8)
+        self.up3  = nn.ConvTranspose2d(base_channels * 8,  base_channels * 4, 2, stride=2)
+        self.dec3 = self._block(base_channels * 8, base_channels * 4)
+        self.up2  = nn.ConvTranspose2d(base_channels * 4,  base_channels * 2, 2, stride=2)
+        self.dec2 = self._block(base_channels * 4, base_channels * 2)
+        self.up1  = nn.ConvTranspose2d(base_channels * 2,  base_channels,     2, stride=2)
+        self.dec1 = self._block(base_channels * 2, base_channels)
+
+        self.out_conv = nn.Conv2d(base_channels, out_channels, kernel_size=1)
+
+    @staticmethod
+    def _block(in_ch: int, out_ch: int) -> nn.Module:
+        return nn.Sequential(
+            nn.Conv2d(in_ch,  out_ch, 3, padding=1, bias=False),
+            nn.BatchNorm2d(out_ch),
+            nn.ReLU(inplace=True),
+            nn.Conv2d(out_ch, out_ch, 3, padding=1, bias=False),
+            nn.BatchNorm2d(out_ch),
+            nn.ReLU(inplace=True),
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        e1 = self.enc1(x)
+        e2 = self.enc2(self.pool(e1))
+        e3 = self.enc3(self.pool(e2))
+        e4 = self.enc4(self.pool(e3))
+        b  = self.bottleneck(self.pool(e4))
+
+        d4 = self.dec4(torch.cat([self.up4(b),  e4], dim=1))
+        d3 = self.dec3(torch.cat([self.up3(d4), e3], dim=1))
+        d2 = self.dec2(torch.cat([self.up2(d3), e2], dim=1))
+        d1 = self.dec1(torch.cat([self.up1(d2), e1], dim=1))
+        return self.out_conv(d1)
+
+
+# ---------------------------------------------------------------------------
+# Segmenter wrapper
+# ---------------------------------------------------------------------------
+
+_IMAGENET_MEAN = (0.485, 0.456, 0.406)
+_IMAGENET_STD  = (0.229, 0.224, 0.225)
+
+
+class UNetSegmenter:
+    """Wraps a UNet with preprocessing, weight loading, inference, and fine-tuning.
+
+    Parameters
+    ----------
+    target_size  : square resolution UNet operates at (default 512)
+    normalize    : "per_image" | "imagenet" | "none"
+    device       : torch device string; defaults to cuda if available
+    """
+
+    def __init__(
+        self,
+        *,
+        target_size: int = 512,
+        normalize:   str = "per_image",
+        device:      str | torch.device | None = None,
+        in_channels: int = 3,
+        base_channels: int = 32,
+        out_channels: int = 2,
+    ):
+        self.target_size = target_size
+        self.normalize   = normalize
+        self.device      = (
+            torch.device(device) if device is not None
+            else torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        )
+        self.model = UNet(in_channels, base_channels, out_channels).to(self.device)
+        self._to_tensor = transforms.ToTensor()
+
+    # ── lifecycle ────────────────────────────────────────────────────────────
+
+    def to(self, device: str | torch.device) -> "UNetSegmenter":
+        self.device = torch.device(device)
+        self.model.to(self.device)
+        return self
+
+    def load_weights(self, path: str | Path) -> "UNetSegmenter":
+        """Load a UNet checkpoint (raw state_dict or {'model': state_dict})."""
+        state = torch.load(Path(path), map_location=self.device, weights_only=False)
+        sd    = state["model"] if isinstance(state, dict) and "model" in state else state
+        self.model.load_state_dict(sd)
+        self.model.eval()
+        return self
+
+    # ── preprocessing ────────────────────────────────────────────────────────
+
+    def _normalize_tensor(self, t: torch.Tensor) -> torch.Tensor:
+        if self.normalize == "per_image":
+            mean = t.mean(dim=(-2, -1), keepdim=True)
+            std  = t.std (dim=(-2, -1), keepdim=True).clamp(min=1e-6)
+            return (t - mean) / std
+        if self.normalize == "imagenet":
+            mean = torch.tensor(_IMAGENET_MEAN, device=t.device).view(-1, 1, 1)
+            std  = torch.tensor(_IMAGENET_STD,  device=t.device).view(-1, 1, 1)
+            return (t - mean) / std
+        return t
+
+    def preprocess(self, image: Image.Image) -> torch.Tensor:
+        """PIL image → normalized (C, H, W) tensor on segmenter device."""
+        resized = image.convert("RGB").resize(
+            (self.target_size, self.target_size), Resampling.BILINEAR
+        )
+        return self._normalize_tensor(self._to_tensor(resized).to(self.device))
+
+    # ── inference ────────────────────────────────────────────────────────────
+
+    @torch.no_grad()
+    def predict(
+        self,
+        image:     Image.Image,
+        *,
+        threshold: float = 0.5,
+        tta:       bool  = False,
+    ) -> tuple[np.ndarray, np.ndarray]:
+        """Single image → (disc_mask, cup_mask) binary uint8 arrays at target_size.
+
+        cup_mask is restricted to disc area (cup ⊆ disc).
+        """
+        self.model.eval()
+        x      = self.preprocess(image).unsqueeze(0)
+        logits = self.model(x)
+        if tta:
+            log_h  = torch.flip(self.model(torch.flip(x, dims=[3])), dims=[3])
+            log_v  = torch.flip(self.model(torch.flip(x, dims=[2])), dims=[2])
+            logits = (logits + log_h + log_v) / 3.0
+        probs = torch.sigmoid(logits)[0].cpu().numpy()
+        disc  = (probs[0] > threshold).astype(np.uint8)
+        cup   = ((probs[1] > threshold) & (disc > 0)).astype(np.uint8)
+        return disc, cup
+
+    # ── fine-tuning ──────────────────────────────────────────────────────────
+
+    def finetune(
+        self,
+        dataloader,
+        *,
+        epochs:     int   = 10,
+        lr:         float = 1e-5,
+        log_prefix: str   = "[UNetSegmenter]",
+    ) -> "UNetSegmenter":
+        """Fine-tune on (image_tensor, mask_tensor) pairs.
+
+        image_tensor : (B, C, H, W) — already preprocessed (normalized)
+        mask_tensor  : (B, 2, H, W) float32 — channel 0 disc, channel 1 cup
+        """
+        import time
+        opt  = torch.optim.Adam(self.model.parameters(), lr=lr)
+        crit = nn.BCEWithLogitsLoss()
+        for ep in range(1, epochs + 1):
+            self.model.train()
+            running, n_batches, t0 = 0.0, 0, time.time()
+            for img, mask in dataloader:
+                img, mask = img.to(self.device), mask.to(self.device)
+                opt.zero_grad()
+                loss = crit(self.model(img), mask)
+                loss.backward()
+                opt.step()
+                running   += float(loss.item())
+                n_batches += 1
+            avg = running / max(n_batches, 1)
+            print(
+                f"  {log_prefix} ep{ep:03d}/{epochs:03d}  loss={avg:.4f}  "
+                f"({time.time() - t0:.1f}s)",
+                flush=True,
+            )
+        self.model.eval()
+        return self

v4/classes/bridges/mono_bridge.py

@@ -0,0 +1,46 @@
+"""mono_bridge — MonoBridge: passthrough for single-tower fusion stages.
+
+The v4 stage runner always expects tower → bridge → head.  For configs that
+have only one tower feeding a head, MonoBridge is the no-op bridge that lets
+the architecture be "head sits directly on tower" without any extra projection,
+SE, or fusion logic.
+
+Optional LayerNorm is exposed for consistency with FusionBridge but defaults
+off to keep the embedding numerically identical to the tower's output.
+"""
+from __future__ import annotations
+
+import torch
+from torch import nn
+
+
+class MonoBridge(nn.Module):
+    """Single-input passthrough bridge.
+
+    Parameters
+    ----------
+    input_dims : list[int] — must be length 1
+    use_ln     : if True, wrap the embedding in a LayerNorm
+    """
+
+    def __init__(self, input_dims: list[int], use_ln: bool = False):
+        super().__init__()
+        if len(input_dims) != 1:
+            raise ValueError(
+                f"MonoBridge expects exactly 1 input dim, got {len(input_dims)}"
+            )
+        self.out_dim = input_dims[0]
+        self.ln      = nn.LayerNorm(self.out_dim) if use_ln else nn.Identity()
+
+    def forward(self, embeddings: list[torch.Tensor]) -> torch.Tensor:
+        if len(embeddings) != 1:
+            raise ValueError(
+                f"MonoBridge forward expects 1 embedding, got {len(embeddings)}"
+            )
+        return self.ln(embeddings[0])
+
+    def set_phase(self, phase: str) -> None:
+        """Freeze during tower_warmup; trainable otherwise (matches FusionBridge)."""
+        enabled = phase not in ("tower_warmup", "cd_warmup")
+        for p in self.parameters():
+            p.requires_grad_(enabled)

v4/classes/logging/prediction_store.py

@@ -205,7 +205,7 @@ class PredictionStore:
                 grp.create_dataset("y_true",     data=buf.y_true)
                 grp.create_dataset("loss",       data=buf.loss)
                 grp.create_dataset("head_names", data=np.array(buf.head_names, dtype=object), dtype=_STR_DT)
-                grp.create_dataset("split",      data=buf.split.astype(str),                  dtype=_STR_DT)
+                grp.create_dataset("split",      data=buf.split,                              dtype=_STR_DT)
                 _write_entity_ids(grp, buf.entity_ids)
 
     @classmethod
@@ -355,7 +355,7 @@ class FeatureStore:
             for phase, buf in self._phases.items():
                 grp = f.create_group(phase)
                 grp.create_dataset("y_true", data=buf.y_true)
-                grp.create_dataset("split",  data=buf.split.astype(str), dtype=_STR_DT)
+                grp.create_dataset("split",  data=buf.split,             dtype=_STR_DT)
                 _write_entity_ids(grp, buf.entity_ids)
                 for head, (arr, _) in buf._heads.items():
                     grp.create_dataset(head, data=arr, compression="gzip", compression_opts=4)

v4/classes/profiles/fundus_images.py

@@ -0,0 +1,574 @@
+"""fundus_images — disc/cup geometry for fundus image profiles.
+
+Contains all fundus-specific geometry logic: mask parsing, feature computation,
+and source-specific loaders.  Profiles whose ImageDataView supports geometry
+should implement build_geometry_loader(source, **kwargs) and/or
+build_seg_map_loader(source, **kwargs) and delegate here.
+
+Geometry vectors (5 scalar CDR features per eye)
+-----------------------------------------------
+  build_geometry_loader("gt", contour_dir=...)         → GTGeometryLoader
+
+Seg maps (3-class disc/cup label map per eye, fed to a CNN tower)
+-----------------------------------------------------------------
+  build_seg_map_loader("gt",   contour_dir=..., ...)   → GTSegMapLoader
+  build_seg_map_loader("unet", weights_path=..., ...)  → UNetSegMapLoader
+"""
+from __future__ import annotations
+
+import copy
+from collections import Counter
+from pathlib import Path
+from typing import Iterable, Tuple
+
+import numpy as np
+import torch
+from PIL import Image, ImageDraw
+from PIL.Image import Resampling
+from torch.utils.data import DataLoader, Dataset
+
+EPS = 1e-6
+
+_FEATURE_DIM   = 5
+_FEATURE_NAMES = ["area_cdr", "rim_ratio", "vertical_cdr", "horizontal_cdr", "centre_shift"]
+
+_MASK_SIZE = (512, 512)  # canonical rasterisation size; CDR ratios are scale-invariant
+
+
+# ---------------------------------------------------------------------------
+# Mask utilities
+# ---------------------------------------------------------------------------
+
+def disc_cup_from_mask_image(mask_img: Image.Image) -> Tuple[np.ndarray, np.ndarray]:
+    """Return binary (disc, cup) masks from a REFUGE-style colour annotation image."""
+    arr = np.asarray(mask_img)
+    if arr.ndim == 3:
+        h, w, c = arr.shape
+        border   = np.concatenate(
+            [arr[0, :, :], arr[-1, :, :], arr[:, 0, :], arr[:, -1, :]], axis=0
+        )
+        bg_color = Counter(map(tuple, border)).most_common(1)[0][0]
+        colors   = Counter(map(tuple, arr.reshape(-1, c)))
+        colors.pop(bg_color, None)
+        disc = (~np.all(arr == bg_color, axis=-1)).astype(np.uint8)
+        if colors:
+            cup_color = min(colors.keys(), key=lambda col: sum(col))
+            cup = np.all(arr == cup_color, axis=-1).astype(np.uint8)
+        else:
+            cup = np.zeros((h, w), dtype=np.uint8)
+    else:
+        border   = np.concatenate([arr[0, :], arr[-1, :], arr[:, 0], arr[:, -1]])
+        bg_value = Counter(border.tolist()).most_common(1)[0][0]
+        disc = (arr != bg_value).astype(np.uint8)
+        fg   = arr[arr != bg_value]
+        cup  = (arr == int(np.min(fg))).astype(np.uint8) if fg.size > 0 else np.zeros_like(arr)
+    cup = (cup > 0) & (disc > 0)
+    return disc.astype(np.uint8), cup.astype(np.uint8)
+
+
+def _contour_to_mask(coords: np.ndarray, size: Tuple[int, int]) -> np.ndarray:
+    """Rasterize a polygon contour (Nx2 xy array) into a binary mask of (width, height)."""
+    from PIL import ImageDraw
+    if coords is None or coords.size == 0:
+        return np.zeros((size[1], size[0]), dtype=np.uint8)
+    img  = Image.new("L", size, 0)
+    draw = ImageDraw.Draw(img)
+    draw.polygon([tuple(map(float, pt)) for pt in coords], outline=1, fill=1)
+    return np.array(img, dtype=np.uint8)
+
+
+# ---------------------------------------------------------------------------
+# Feature computation
+# ---------------------------------------------------------------------------
+
+def compute_geometry_features(disc_mask: np.ndarray, cup_mask: np.ndarray) -> np.ndarray:
+    """Compute 5 cup/disc structural descriptors from binary masks.
+
+    Returns float32 [area_cdr, rim_ratio, vertical_cdr, horizontal_cdr, centre_shift].
+    """
+    disc = (disc_mask > 0).astype(np.float32)
+    cup  = (cup_mask  > 0).astype(np.float32)
+
+    disc_area = disc.sum()
+    cup_area  = cup.sum()
+    area_cdr  = cup_area  / (disc_area + EPS)
+    rim_ratio = (disc_area - cup_area) / (disc_area + EPS)
+
+    disc_h = float(np.any(disc > 0, axis=1).sum())
+    cup_h  = float(np.any(cup  > 0, axis=1).sum())
+    disc_w = float(np.any(disc > 0, axis=0).sum())
+    cup_w  = float(np.any(cup  > 0, axis=0).sum())
+
+    vertical_cdr   = cup_h / (disc_h + EPS)
+    horizontal_cdr = cup_w / (disc_w + EPS)
+
+    def _centre(m: np.ndarray) -> Tuple[float, float]:
+        coords = np.argwhere(m > 0)
+        if coords.size == 0:
+            return 0.5, 0.5
+        ys, xs = coords[:, 0], coords[:, 1]
+        return float(xs.mean()) / m.shape[1], float(ys.mean()) / m.shape[0]
+
+    dcx, dcy = _centre(disc)
+    ccx, ccy = _centre(cup)
+    centre_shift = float(np.hypot(ccx - dcx, ccy - dcy))
+
+    return np.array(
+        [area_cdr, rim_ratio, vertical_cdr, horizontal_cdr, centre_shift],
+        dtype=np.float32,
+    )
+
+
+# ---------------------------------------------------------------------------
+# Loaders
+# ---------------------------------------------------------------------------
+
+class GTGeometryLoader:
+    """Pre-computes per-eye geometry vectors from PAPILA GT contour annotations.
+
+    File naming: RET{pid:03d}{eye}_{disc|cup}_exp{n}.txt
+    Averages exp1 and exp2 when both are present; zero vector for missing entries.
+
+    Usage:
+        loader = GTGeometryLoader(contour_dir)
+        loader.precompute(df, patient_col="Patient ID")
+        vecs = loader.all_vectors()   # {(pid, eye): ndarray}
+    """
+
+    _EXPERTS = (1, 2)
+
+    feature_dim   = _FEATURE_DIM
+    feature_names = _FEATURE_NAMES
+
+    def __init__(self, contour_dir: str | Path) -> None:
+        self._dir   = Path(contour_dir)
+        self._cache: dict[tuple, np.ndarray] = {}
+
+    def precompute(self, df, patient_col: str = "Patient ID") -> None:
+        n_ok = 0
+        for _, row in df.iterrows():
+            pid = int(row[patient_col])
+            eye = str(row.get("eyeID", "OD"))
+            key = (pid, eye)
+            if key in self._cache:
+                continue
+            vec = self._compute(pid, eye)
+            self._cache[key] = vec if vec is not None else np.zeros(self.feature_dim, dtype=np.float32)
+            if vec is not None:
+                n_ok += 1
+        print(f"[GTGeometryLoader] {n_ok}/{len(self._cache)} geometry vectors computed", flush=True)
+
+    def all_vectors(self) -> dict:
+        return dict(self._cache)
+
+    def _compute(self, pid: int, eye: str) -> "np.ndarray | None":
+        stem = f"RET{pid:03d}{eye}"
+        vecs: list[np.ndarray] = []
+        for exp in self._EXPERTS:
+            disc_path = self._dir / f"{stem}_disc_exp{exp}.txt"
+            cup_path  = self._dir / f"{stem}_cup_exp{exp}.txt"
+            if not disc_path.exists():
+                continue
+            try:
+                disc_c = np.loadtxt(disc_path)
+                if disc_c.ndim == 1:
+                    disc_c = disc_c.reshape(-1, 2)
+                disc_mask = _contour_to_mask(disc_c, _MASK_SIZE)
+                if cup_path.exists():
+                    cup_c = np.loadtxt(cup_path)
+                    if cup_c.ndim == 1:
+                        cup_c = cup_c.reshape(-1, 2)
+                    cup_mask = _contour_to_mask(cup_c, _MASK_SIZE)
+                else:
+                    cup_mask = np.zeros((_MASK_SIZE[1], _MASK_SIZE[0]), dtype=np.uint8)
+                cup_mask = ((cup_mask > 0) & (disc_mask > 0)).astype(np.uint8)
+                vecs.append(compute_geometry_features(disc_mask, cup_mask))
+            except Exception:
+                continue
+        if not vecs:
+            return None
+        return np.stack(vecs).mean(axis=0).astype(np.float32)
+
+
+# ---------------------------------------------------------------------------
+# Seg-map utilities (shared by GT and UNet seg-map loaders)
+# ---------------------------------------------------------------------------
+
+def _combine_disc_cup(disc: np.ndarray, cup: np.ndarray) -> np.ndarray:
+    """Merge binary disc + cup masks into a uint8 label map: 0=bg, 1=rim, 2=cup."""
+    disc = (disc > 0).astype(np.uint8)
+    cup  = ((cup > 0) & (disc > 0)).astype(np.uint8)
+    return (disc + cup).astype(np.uint8)
+
+
+def _crop_to_disc_bbox(seg_map: np.ndarray) -> np.ndarray:
+    """Crop a label map tightly to the disc bounding box (anywhere seg_map > 0)."""
+    rows = np.any(seg_map > 0, axis=1)
+    cols = np.any(seg_map > 0, axis=0)
+    if not rows.any():
+        return seg_map
+    r0, r1 = int(np.argmax(rows)), int(len(rows) - 1 - np.argmax(rows[::-1]))
+    c0, c1 = int(np.argmax(cols)), int(len(cols) - 1 - np.argmax(cols[::-1]))
+    return seg_map[r0:r1 + 1, c0:c1 + 1]
+
+
+def _seg_map_to_array(
+    seg_map: np.ndarray, channels: int, target_size: int
+) -> np.ndarray:
+    """Resize a {0,1,2} seg map and convert to a (C, H, W) float32 array.
+
+    channels=1  →  (1, H, W) values in {0, 0.5, 1.0}
+    channels=3  →  (3, H, W) one-hot [bg, rim, cup]
+    """
+    pil = Image.fromarray(seg_map.astype(np.uint8), mode="L").resize(
+        (target_size, target_size), Resampling.NEAREST
+    )
+    arr = np.array(pil, dtype=np.uint8)
+    if channels == 1:
+        return (arr.astype(np.float32) / 2.0)[None, :, :]
+    if channels == 3:
+        return np.stack([
+            (arr == 0).astype(np.float32),
+            (arr == 1).astype(np.float32),
+            (arr == 2).astype(np.float32),
+        ], axis=0)
+    raise ValueError(f"channels must be 1 or 3, got {channels}")
+
+
+def _load_papila_contour(path: Path) -> np.ndarray:
+    """Load (x, y) contour pairs from a PAPILA whitespace/comma-delimited text file."""
+    for delim in (",", None):
+        try:
+            arr = np.loadtxt(str(path), delimiter=delim, comments="#", dtype=np.float32)
+            if arr.size > 0 and arr.ndim >= 1:
+                if arr.ndim == 1:
+                    arr = arr.reshape(-1, 2)
+                if arr.shape[1] >= 2:
+                    return arr[:, :2]
+        except Exception:
+            continue
+    return np.zeros((0, 2), dtype=np.float32)
+
+
+def _papila_disc_cup_masks(
+    pid: int, eye: str, contour_dir: Path, image_size: Tuple[int, int],
+    mask_size: int, experts: Tuple[int, ...] = (1, 2),
+) -> Tuple[np.ndarray, np.ndarray] | None:
+    """Average masks across PAPILA experts. Returns (disc, cup) at mask_size, or None."""
+    stem = f"RET{pid:03d}{eye}"
+    discs, cups = [], []
+    for exp in experts:
+        disc_path = contour_dir / f"{stem}_disc_exp{exp}.txt"
+        cup_path  = contour_dir / f"{stem}_cup_exp{exp}.txt"
+        if not disc_path.exists():
+            continue
+        disc_c = _load_papila_contour(disc_path)
+        if len(disc_c) < 3:
+            continue
+        disc_m = _rasterise_polygon(disc_c, image_size, mask_size)
+        if cup_path.exists():
+            cup_c = _load_papila_contour(cup_path)
+            cup_m = (_rasterise_polygon(cup_c, image_size, mask_size)
+                     if len(cup_c) >= 3 else np.zeros_like(disc_m))
+        else:
+            cup_m = np.zeros_like(disc_m)
+        discs.append(disc_m)
+        cups.append(cup_m)
+    if not discs:
+        return None
+    disc = (np.mean(discs, axis=0) > 0.5).astype(np.uint8)
+    cup  = (np.mean(cups,  axis=0) > 0.5).astype(np.uint8)
+    return disc, cup
+
+
+def _rasterise_polygon(
+    coords: np.ndarray, image_size: Tuple[int, int], target_size: int
+) -> np.ndarray:
+    """Rasterise an (N, 2) polygon contour into a (target_size, target_size) binary mask.
+
+    image_size is (width, height) of the coord space (the original fundus image).
+    """
+    if coords is None or len(coords) < 3:
+        return np.zeros((target_size, target_size), dtype=np.uint8)
+    img  = Image.new("L", image_size, 0)
+    pts  = [tuple(map(float, p)) for p in coords]
+    ImageDraw.Draw(img).polygon(pts, outline=1, fill=1)
+    img  = img.resize((target_size, target_size), Resampling.NEAREST)
+    return (np.array(img, dtype=np.uint8) > 0).astype(np.uint8)
+
+
+# ---------------------------------------------------------------------------
+# Seg-map loaders
+# ---------------------------------------------------------------------------
+
+class GTSegMapLoader:
+    """Pre-computes per-eye disc/cup seg maps from PAPILA GT contour annotations.
+
+    Output: dict {(pid, eye): np.ndarray (C, H, W) float32} cached for the fold.
+    """
+
+    def __init__(
+        self,
+        contour_dir: str | Path,
+        *,
+        channels:     int  = 3,
+        mask_size:    int  = 512,
+        target_size:  int  = 224,
+        crop_to_disc: bool = True,
+    ) -> None:
+        self._dir         = Path(contour_dir)
+        self._channels    = channels
+        self._mask_size   = mask_size
+        self._target_size = target_size
+        self._crop        = crop_to_disc
+        self._cache: dict[tuple, np.ndarray] = {}
+
+    @property
+    def cache_dim(self) -> tuple[int, int, int]:
+        return (self._channels, self._target_size, self._target_size)
+
+    def precompute(self, samples: Iterable[Tuple[int, str, Path]]) -> None:
+        """samples: iterable of (pid, eye, image_path) tuples."""
+        n_ok = 0
+        blank = np.zeros((self._mask_size, self._mask_size), dtype=np.uint8)
+        for pid, eye, image_path in samples:
+            key = (pid, eye)
+            if key in self._cache:
+                continue
+            with Image.open(image_path) as _im:
+                im_size = _im.size  # (W, H)
+            res = _papila_disc_cup_masks(
+                pid, eye, self._dir, im_size, self._mask_size,
+            )
+            if res is None:
+                seg = blank
+            else:
+                disc, cup = res
+                seg = _combine_disc_cup(disc, cup)
+                n_ok += 1
+            if self._crop:
+                seg = _crop_to_disc_bbox(seg)
+            self._cache[key] = _seg_map_to_array(seg, self._channels, self._target_size)
+        print(
+            f"[GTSegMapLoader] {n_ok}/{len(self._cache)} GT seg maps computed "
+            f"(channels={self._channels}, target={self._target_size})",
+            flush=True,
+        )
+
+    def all_seg_maps(self) -> dict:
+        return self._cache
+
+
+class _UNetFTDataset(Dataset):
+    """Pre-cached (image_tensor, mask_tensor) pairs for fine-tuning a UNet.
+
+    Decode + resize + normalize + GT mask rasterisation are all deterministic,
+    so we do them once at construction and store float32 tensors on CPU.  This
+    drops per-batch cost to a tensor lookup + GPU transfer.
+    """
+
+    def __init__(
+        self,
+        records:     list,         # list of (pid, eye, image_path)
+        contour_dir: Path,
+        segmenter,                 # UNetSegmenter — used for image preprocessing
+    ):
+        import time
+        S = segmenter.target_size
+        self._imgs:  list[torch.Tensor] = []
+        self._masks: list[torch.Tensor] = []
+
+        print(
+            f"[UNetSegMapLoader] pre-caching {len(records)} (image, mask) pairs "
+            f"at {S}×{S}...",
+            flush=True,
+        )
+        t0     = time.time()
+        report = max(1, len(records) // 4)
+        for i, (pid, eye, image_path) in enumerate(records, 1):
+            with Image.open(image_path) as raw:
+                im_size = raw.size
+                img_t   = segmenter.preprocess(raw).detach().cpu()
+            res = _papila_disc_cup_masks(pid, eye, contour_dir, im_size, S)
+            if res is None:
+                disc = np.zeros((S, S), dtype=np.uint8)
+                cup  = np.zeros_like(disc)
+            else:
+                disc, cup = res
+            mask_t = torch.from_numpy(np.stack([disc, cup], axis=0).astype(np.float32))
+            self._imgs.append(img_t)
+            self._masks.append(mask_t)
+            if i % report == 0 or i == len(records):
+                print(
+                    f"  [UNet ft cache] {i}/{len(records)}  ({time.time() - t0:.1f}s)",
+                    flush=True,
+                )
+
+    def __len__(self) -> int:
+        return len(self._imgs)
+
+    def __getitem__(self, idx: int):
+        return self._imgs[idx], self._masks[idx]
+
+
+class UNetSegMapLoader:
+    """Pre-computes per-eye seg maps via a REFUGE-pretrained UNet.
+
+    Optionally fine-tunes the UNet per fold on the training split's GT contours.
+
+    Output: dict {(pid, eye): np.ndarray (C, H, W) float32} cached for the fold.
+    """
+
+    def __init__(
+        self,
+        weights_path: str | Path,
+        *,
+        contour_dir:         str | Path,
+        channels:            int   = 3,
+        target_size:         int   = 224,
+        unet_size:           int   = 512,
+        normalize:           str   = "per_image",
+        threshold:           float = 0.5,
+        crop_to_disc:        bool  = True,
+        finetune_epochs:     int   = 0,
+        finetune_lr:         float = 1e-5,
+        finetune_batch_size: int   = 4,
+        device:              str | None = None,
+    ) -> None:
+        from v4.classes.accessory.unet import UNetSegmenter
+
+        self._weights_path        = Path(weights_path)
+        self._contour_dir         = Path(contour_dir)
+        self._channels            = channels
+        self._target_size         = target_size
+        self._threshold           = threshold
+        self._crop                = crop_to_disc
+        self._ft_epochs           = finetune_epochs
+        self._ft_lr               = finetune_lr
+        self._ft_batch_size       = finetune_batch_size
+
+        self._segmenter = UNetSegmenter(
+            target_size=unet_size, normalize=normalize, device=device,
+        ).load_weights(self._weights_path)
+        self._base_state = copy.deepcopy(self._segmenter.model.state_dict())
+        self._cache: dict[tuple, np.ndarray] = {}
+
+    @property
+    def cache_dim(self) -> tuple[int, int, int]:
+        return (self._channels, self._target_size, self._target_size)
+
+    def reset_cache(self) -> None:
+        """Clear cached seg maps (call between folds)."""
+        self._cache.clear()
+
+    def reset_weights(self) -> None:
+        """Restore base REFUGE weights (undo any prior fine-tuning)."""
+        self._segmenter.model.load_state_dict(copy.deepcopy(self._base_state))
+
+    def finetune(self, train_samples: list) -> None:
+        """Fine-tune the UNet on the training fold's GT contours.
+
+        train_samples: list of (pid, eye, image_path) tuples — train split only.
+        """
+        if self._ft_epochs <= 0:
+            return
+        ds = _UNetFTDataset(train_samples, self._contour_dir, self._segmenter)
+        loader = DataLoader(
+            ds, batch_size=self._ft_batch_size, shuffle=True, num_workers=0,
+        )
+        print(
+            f"[UNetSegMapLoader] fine-tuning UNet for {self._ft_epochs} epochs "
+            f"on {len(train_samples)} samples (lr={self._ft_lr}, "
+            f"batch_size={self._ft_batch_size})",
+            flush=True,
+        )
+        self._segmenter.finetune(
+            loader, epochs=self._ft_epochs, lr=self._ft_lr,
+            log_prefix="[UNet ft]",
+        )
+
+    def precompute(self, samples: Iterable[Tuple[int, str, Path]]) -> None:
+        """Run UNet inference on every sample and cache the resulting seg map."""
+        import time
+        samples = list(samples)
+        todo    = [s for s in samples if (s[0], s[1]) not in self._cache]
+        if not todo:
+            return
+        print(
+            f"[UNetSegMapLoader] running UNet inference on {len(todo)} images...",
+            flush=True,
+        )
+        t0     = time.time()
+        report = max(1, len(todo) // 4)
+        for i, (pid, eye, image_path) in enumerate(todo, 1):
+            with Image.open(image_path) as raw:
+                disc, cup = self._segmenter.predict(raw, threshold=self._threshold)
+            seg = _combine_disc_cup(disc, cup)
+            if self._crop:
+                seg = _crop_to_disc_bbox(seg)
+            self._cache[(pid, eye)] = _seg_map_to_array(
+                seg, self._channels, self._target_size,
+            )
+            if i % report == 0 or i == len(todo):
+                print(
+                    f"  [UNet inf] {i}/{len(todo)}  ({time.time() - t0:.1f}s)",
+                    flush=True,
+                )
+        print(
+            f"[UNetSegMapLoader] {len(todo)} seg maps cached via UNet "
+            f"(channels={self._channels}, target={self._target_size})",
+            flush=True,
+        )
+
+    def all_seg_maps(self) -> dict:
+        return self._cache
+
+
+# ---------------------------------------------------------------------------
+# Factories
+# ---------------------------------------------------------------------------
+
+def build_geometry_loader(source: str, **kwargs):
+    """Return the appropriate geometry-vector loader for the given source string.
+
+    Parameters
+    ----------
+    source      : "gt" | "unet"
+    contour_dir : (gt) path to contour annotation directory
+    """
+    if source == "gt":
+        contour_dir = kwargs.get("contour_dir")
+        if contour_dir is None:
+            raise ValueError("build_geometry_loader source='gt' requires contour_dir")
+        return GTGeometryLoader(contour_dir)
+    raise NotImplementedError(f"build_geometry_loader: source={source!r} not implemented")
+
+
+def build_seg_map_loader(source: str, **kwargs):
+    """Return the appropriate seg-map loader for the given source string.
+
+    Parameters
+    ----------
+    source : "gt" | "unet"
+
+    GT kwargs:
+        contour_dir, channels=3, mask_size=512, target_size=224, crop_to_disc=True
+    UNet kwargs:
+        weights_path, contour_dir, channels=3, target_size=224, unet_size=512,
+        normalize="per_image", threshold=0.5, crop_to_disc=True,
+        finetune_epochs=0, finetune_lr=1e-5, finetune_batch_size=4, device=None
+    """
+    if source == "gt":
+        if "contour_dir" not in kwargs:
+            raise ValueError("build_seg_map_loader source='gt' requires contour_dir")
+        return GTSegMapLoader(**kwargs)
+    if source == "unet":
+        if "weights_path" not in kwargs:
+            raise ValueError("build_seg_map_loader source='unet' requires weights_path")
+        if "contour_dir" not in kwargs:
+            raise ValueError(
+                "build_seg_map_loader source='unet' requires contour_dir "
+                "(needed for per-fold fine-tuning, even if finetune_epochs=0)"
+            )
+        return UNetSegMapLoader(**kwargs)
+    raise NotImplementedError(f"build_seg_map_loader: source={source!r} not implemented")

v4/classes/profiles/v4papila.py

@@ -418,6 +418,30 @@ class ImageDataView:
             eye_filter=eye,
         )
 
+    # ── Geometry hook ─────────────────────────────────────────────────────────
+
+    def _resolve_paths(self, kwargs: dict) -> dict:
+        """Resolve any *_dir / *_path kwargs against the repo root."""
+        repo_root = Path(__file__).resolve().parents[3]
+        out = {}
+        for k, v in kwargs.items():
+            if (k.endswith("_dir") or k.endswith("_path")) and v is not None:
+                p = Path(v)
+                out[k] = str(repo_root / p) if not p.is_absolute() else v
+            else:
+                out[k] = v
+        return out
+
+    def build_geometry_loader(self, source: str, **kwargs):
+        """Return a geometry-vector loader (delegates to fundus_images)."""
+        from v4.classes.profiles.fundus_images import build_geometry_loader as _build
+        return _build(source, **self._resolve_paths(kwargs))
+
+    def build_seg_map_loader(self, source: str, **kwargs):
+        """Return a seg-map loader (delegates to fundus_images)."""
+        from v4.classes.profiles.fundus_images import build_seg_map_loader as _build
+        return _build(source, **self._resolve_paths(kwargs))
+
 
 # ---------------------------------------------------------------------------
 # PapilaBundle — the v4 DataBundle returned by build_data
@@ -515,6 +539,7 @@ class PapilaBundle:
         *,
         level: str = "eye",
         label_filter: list[int] | None = None,
+        eye_filter: str | None = None,
     ) -> LoaderShell:
         """Build a LoaderShell from a split DataFrame.
 
@@ -533,6 +558,8 @@ class PapilaBundle:
 
         if label_filter is not None:
             df = df[df[lc].isin(label_filter)]
+        if eye_filter is not None and "eyeID" in df.columns:
+            df = df[df["eyeID"] == eye_filter]
 
         entries: list[ShellEntry] = []
 

v4/classes/stages/fusion.py

@@ -11,7 +11,8 @@ import torch.nn.functional as F
 from v4.classes.dataset import LoaderShell, to_label_tensor
 from v4.classes.metrics import score_arrays, compute_extended_metrics, tune_binary_threshold
 from v4.classes.stages.helpers import (
-    encode_embedding, get_out_dim, resolve_input_dims, phase_for_epoch,
+    class_weights_from_shell, encode_embedding, get_out_dim, resolve_input_dims,
+    phase_for_epoch,
 )
 
 
@@ -25,15 +26,15 @@ def collect_probs(
     loader,
     device,
     num_classes:  int,
-) -> tuple[np.ndarray, np.ndarray]:
-    """Eval pass for one fusion stage; returns (y_true, softmax_probs)."""
+) -> tuple[np.ndarray, np.ndarray, list, np.ndarray]:
+    """Eval pass for one fusion stage; returns (y_true, softmax_probs, entity_ids, embeddings)."""
     from v4.classes.dataset import to_label_tensor
     bridge.eval(); primary_head.eval()
     for t in towers.values():
         t.eval()
     inputs       = stage_cfg["inputs"]
     is_bilateral = isinstance(inputs, dict)
-    y_all, p_all = [], []
+    y_all, p_all, ids_all, z_all = [], [], [], []
 
     with torch.no_grad():
         for batch in loader:
@@ -57,10 +58,14 @@ def collect_probs(
             logits = primary_head(z)
             y_all.append(to_label_tensor(y, device).cpu().numpy())
             p_all.append(F.softmax(logits, dim=1).cpu().numpy())
+            z_all.append(z.cpu().numpy())
+            ids_all.extend(batch.get("entity_id", []))
 
     if not y_all:
-        return np.zeros(0, dtype=np.int64), np.zeros((0, num_classes), dtype=np.float32)
-    return np.concatenate(y_all), np.concatenate(p_all, axis=0)
+        return (np.zeros(0, dtype=np.int64), np.zeros((0, num_classes), dtype=np.float32),
+                [], np.zeros((0, 0), dtype=np.float32))
+    return (np.concatenate(y_all), np.concatenate(p_all, axis=0),
+            ids_all, np.concatenate(z_all, axis=0))
 
 
 def run(
@@ -88,9 +93,13 @@ def run(
     inputs       = stage_cfg["inputs"]
     is_bilateral = isinstance(inputs, dict)
 
-    s_train = data.build_shells(split.train, level=level, label_filter=label_filter)
-    s_val   = data.build_shells(split.val,   level=level, label_filter=label_filter)
-    s_test  = (data.build_shells(split.test, level=level, label_filter=label_filter)
+    eye_filter = stage_cfg.get("eye_filter", None)
+    s_train = data.build_shells(split.train, level=level, label_filter=label_filter,
+                                eye_filter=eye_filter)
+    s_val   = data.build_shells(split.val,   level=level, label_filter=label_filter,
+                                eye_filter=eye_filter)
+    s_test  = (data.build_shells(split.test, level=level, label_filter=label_filter,
+                                 eye_filter=eye_filter)
                if split.test is not None else LoaderShell(entries=[]))
 
     if not s_val.entries:
@@ -116,7 +125,11 @@ def run(
         h_dim = get_out_dim(hs["input"], towers, {**stage_models, name: bridge})
         h_mod = importlib.import_module(hs.get("module", "v4.classes.heads.classifier"))
         h_cls = getattr(h_mod, hs.get("class", "ClassificationHead"))
-        head_models[hs["name"]] = h_cls(h_dim, num_classes).to(device)
+        existing = stage_models.get(hs["name"])
+        head_models[hs["name"]] = (
+            existing.to(device) if existing is not None
+            else h_cls(h_dim, num_classes, **hs.get("args", {})).to(device)
+        )
 
     primary_hs_cfg = next((hs for hs in head_stage_cfgs if not hs.get("bcd", False)), None)
     bcd_head_cfgs  = [hs for hs in head_stage_cfgs if hs.get("bcd", False)]
@@ -132,11 +145,23 @@ def run(
         for p in m.parameters():
             p.requires_grad_(False)
         m.eval()
+    for h in head_models.values():
+        for p in h.parameters():
+            p.requires_grad_(True)
 
     # ── Optimizer ────────────────────────────────────────────────────────────
     train_towers = stage_cfg.get("train_towers", False)
+    if train_towers:
+        # Only train towers that are direct inputs to this stage (not all towers globally).
+        # For nt_od with inputs ["img_od", "cd_od"] this trains only those two; other
+        # eye's towers remain untouched.
+        direct_inputs = list(inputs.values()) if isinstance(inputs, dict) else inputs
+        tower_params  = [p for n in direct_inputs if n in towers
+                         for p in towers[n].parameters()]
+    else:
+        tower_params = []
     opt_params = (
-        ([p for t in towers.values() for p in t.parameters()] if train_towers else []) +
+        tower_params +
         list(bridge.parameters()) +
         [p for h in head_models.values() for p in h.parameters()]
     )
@@ -147,6 +172,17 @@ def run(
     wf         = 0 if is_bilateral else warmup_cfg.get("fused_epochs",  0)
     bcd_prob   = cfg["training"].get("bcd_prob", 0.5)
 
+    cw = class_weights_from_shell(
+        s_train, num_classes, device,
+        enabled=cfg["training"].get("class_weighted", False),
+    )
+    if cw is not None:
+        print(
+            f"  fold{fold+1} [{name}] class weights: "
+            + ", ".join(f"{i}={w:.3f}" for i, w in enumerate(cw.tolist())),
+            flush=True,
+        )
+
     # ── Epoch loop ────────────────────────────────────────────────────────────
     for epoch in range(epochs):
         bridge.train()
@@ -200,7 +236,7 @@ def run(
             if is_bilateral or phase == "fused_warmup":
                 logits = head_logits.get(primary_hs_cfg["name"])
             elif phase == "tower_warmup" and bcd_head_cfgs:
-                losses = [F.cross_entropy(head_logits[hs["name"]], y_t)
+                losses = [F.cross_entropy(head_logits[hs["name"]], y_t, weight=cw)
                           for hs in bcd_head_cfgs if hs["name"] in head_logits]
                 if not losses:
                     continue
@@ -217,7 +253,7 @@ def run(
 
             if logits is None:
                 continue
-            loss = F.cross_entropy(logits, y_t)
+            loss = F.cross_entropy(logits, y_t, weight=cw)
             opt.zero_grad(); loss.backward(); opt.step()
             total_correct += int((logits.argmax(1) == y_t).sum())
             total_loss    += loss.item() * len(y_t)
@@ -226,7 +262,7 @@ def run(
         tr_loss = total_loss / total_n if total_n else nan
         tr_acc  = total_correct / total_n if total_n else nan
 
-        y_v, p_v = collect_probs(bridge, primary_head, stage_cfg, towers,
+        y_v, p_v, _, _ = collect_probs(bridge, primary_head, stage_cfg, towers,
                                        stage_models, cfg_stages, val_loader, device, num_classes)
         _, val_auc, _ = score_arrays(y_v, p_v, num_classes) if y_v.size else (nan, nan, nan)
         print(
@@ -236,7 +272,7 @@ def run(
         )
 
     # ── Final eval ────────────────────────────────────────────────────────────
-    y_val, p_val = collect_probs(bridge, primary_head, stage_cfg, towers,
+    y_val, p_val, ids_val, z_val = collect_probs(bridge, primary_head, stage_cfg, towers,
                                                   stage_models, cfg_stages, val_loader, device, num_classes)
     val_acc, val_auc, val_n = (score_arrays(y_val, p_val, num_classes)
                                if y_val.size else (nan, nan, nan))
@@ -247,9 +283,10 @@ def run(
             and num_classes == 2 and y_val.size >= 2):
         val_threshold = tune_binary_threshold(y_val, p_val[:, 1])
 
+    y_te = p_te = ids_te = z_te = None
     test_auc = test_acc = test_n = nan
     if test_loader is not None:
-        y_te, p_te = collect_probs(bridge, primary_head, stage_cfg, towers,
+        y_te, p_te, ids_te, z_te = collect_probs(bridge, primary_head, stage_cfg, towers,
                                                   stage_models, cfg_stages, test_loader, device, num_classes)
         test_acc, test_auc, test_n = (score_arrays(y_te, p_te, num_classes)
                                       if y_te.size else (nan, nan, nan))
@@ -270,4 +307,10 @@ def run(
         f"{name}_test_acc":      test_acc,
         f"{name}_test_n":        test_n,
     }
-    return updated, metrics
+    pred_data = {
+        name: {
+            "val_y": y_val, "val_p": p_val, "val_ids": ids_val, "val_z": z_val,
+            "test_y": y_te, "test_p": p_te, "test_ids": ids_te, "test_z": z_te,
+        }
+    }
+    return updated, metrics, pred_data

v4/classes/stages/helpers.py

@@ -1,9 +1,32 @@
 """stages/helpers — shared utilities for stage runners."""
 from __future__ import annotations
 
+from collections import Counter
+
 import torch
 
 
+def class_weights_from_shell(
+    shell, num_classes: int, device, *, enabled: bool = True
+) -> torch.Tensor | None:
+    """Return inverse-frequency CE weights normalised to mean 1, or None.
+
+    weights[i] = (N_total / num_classes) / N_class_i  → rare classes weighted higher.
+    Mean(weights) ≈ 1 so overall loss magnitude is unchanged.
+
+    Classes absent from the shell get weight 1.0 (no division-by-zero).
+    """
+    if not enabled or shell is None or not shell.entries:
+        return None
+    counts  = Counter(int(e.label) for e in shell.entries)
+    n_total = sum(counts.values())
+    weights = []
+    for c in range(num_classes):
+        n_c = counts.get(c, 0)
+        weights.append(1.0 if n_c == 0 else n_total / (num_classes * n_c))
+    return torch.tensor(weights, dtype=torch.float32, device=device)
+
+
 def get_out_dim(name: str, towers: dict, stage_models: dict) -> int:
     if name in towers:
         return towers[name].out_dim

v4/classes/stages/parallel.py

@@ -0,0 +1,394 @@
+"""stages/parallel — runs multiple same-type sub-stages in a shared epoch loop.
+
+Used to train bilateral pairs (OD + OS) simultaneously rather than sequentially.
+Sub-stages must all be the same type: either all 'warm' or all 'fusion'.
+"""
+from __future__ import annotations
+
+import importlib
+from random import choice, random as _random
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+
+from v4.classes.dataset import LoaderShell, to_label_tensor
+from v4.classes.metrics import score_arrays, compute_extended_metrics, tune_binary_threshold
+from v4.classes.stages.helpers import (
+    class_weights_from_shell, get_out_dim, resolve_input_dims, phase_for_epoch,
+)
+from v4.classes.stages.fusion import collect_probs
+
+
+def run(
+    stage_cfg:        dict,
+    cfg:              dict,
+    towers:           dict,
+    stage_models:     dict,
+    data,
+    split,
+    label_filter,
+    num_classes:      int,
+    device,
+    fold:             int,
+    cfg_stages:       list[dict],
+    _make_loader,
+    _balanced_sampler,
+) -> tuple[dict, dict, dict]:
+    sub_cfgs  = stage_cfg["stages"]
+    sub_types = {s["type"] for s in sub_cfgs}
+
+    if sub_types == {"warm"}:
+        return _parallel_warm(sub_cfgs, cfg, towers, stage_models, data, split,
+                              label_filter, num_classes, device, fold,
+                              cfg_stages, _make_loader, _balanced_sampler)
+    elif sub_types == {"fusion"}:
+        return _parallel_fusion(sub_cfgs, cfg, towers, stage_models, data, split,
+                                label_filter, num_classes, device, fold,
+                                cfg_stages, _make_loader)
+    else:
+        raise ValueError(
+            f"parallel stage sub-stages must all be the same type (warm or fusion), "
+            f"got {sub_types}"
+        )
+
+
+# ── Parallel warm ─────────────────────────────────────────────────────────────
+
+def _parallel_warm(
+    sub_cfgs, cfg, towers, stage_models, data, split,
+    label_filter, num_classes, device, fold, cfg_stages, _make_loader, _balanced_sampler,
+):
+    bs = cfg["training"]["batch_size"]
+
+    contexts = []
+    for sc in sub_cfgs:
+        tower_name   = sc["tower"]
+        level        = sc["level"]
+        shell_filter = sc.get("shell_filter", {})
+        n_epochs     = sc.get("epochs", 0)
+
+        s_train = data.build_shells(split.train, level=level, label_filter=label_filter,
+                                    **shell_filter)
+        loader  = _make_loader(s_train, {tower_name: towers[tower_name]},
+                               batch_size=bs, shuffle=False,
+                               sampler=_balanced_sampler(s_train))
+        head_name = sc.get("head_name")
+        if head_name:
+            head_cfg = next((s for s in cfg_stages if s.get("name") == head_name), None)
+            if head_cfg is None:
+                raise ValueError(f"warm stage requested head_name={head_name!r}, but no such head exists")
+            h_mod = importlib.import_module(head_cfg.get("module", "v4.classes.heads.classifier"))
+            h_cls = getattr(h_mod, head_cfg.get("class", "ClassificationHead"))
+            probe = stage_models.get(head_name)
+            if probe is None:
+                probe = h_cls(towers[tower_name].out_dim, num_classes, **head_cfg.get("args", {}))
+            probe = probe.to(device)
+        else:
+            probe = torch.nn.Linear(towers[tower_name].out_dim, num_classes).to(device)
+        opt   = torch.optim.Adam(
+            list(towers[tower_name].parameters()) + list(probe.parameters()),
+            lr=cfg["training"]["lr"],
+        )
+        cw = class_weights_from_shell(
+            s_train, num_classes, device,
+            enabled=cfg["training"].get("class_weighted", False),
+        )
+        contexts.append({
+            "name": tower_name, "n_epochs": n_epochs,
+            "loader": loader, "probe": probe, "opt": opt,
+            "head_name": head_name, "class_weights": cw,
+        })
+
+    active = {c["name"] for c in contexts if c["n_epochs"] > 0}
+    for name, t in towers.items():
+        for p in t.parameters():
+            p.requires_grad_(name in active)
+
+    max_epochs = max((c["n_epochs"] for c in contexts), default=0)
+    for epoch in range(max_epochs):
+        for ctx in contexts:
+            if epoch >= ctx["n_epochs"]:
+                continue
+            tower_name = ctx["name"]
+            towers[tower_name].train()
+            total_loss = total_correct = total_n = 0
+            for batch in ctx["loader"]:
+                y = batch.get("label")
+                x = batch.get(tower_name)
+                if not torch.is_tensor(y) or not torch.is_tensor(x):
+                    continue
+                y_t    = to_label_tensor(y, device)
+                logits = ctx["probe"](towers[tower_name](x.to(device)))
+                loss   = F.cross_entropy(logits, y_t, weight=ctx["class_weights"])
+                ctx["opt"].zero_grad(); loss.backward(); ctx["opt"].step()
+                total_loss    += loss.item() * len(y_t)
+                total_correct += int((logits.argmax(1) == y_t).sum())
+                total_n       += len(y_t)
+            if total_n:
+                print(
+                    f"  fold{fold+1} [warm/{tower_name}]"
+                    f" ep{epoch+1:03d}/{ctx['n_epochs']}"
+                    f"  loss={total_loss/total_n:.4f}  acc={total_correct/total_n:.3f}",
+                    flush=True,
+                )
+
+    for t in towers.values():
+        for p in t.parameters():
+            p.requires_grad_(True)
+
+    updated = dict(stage_models)
+    for ctx in contexts:
+        if ctx.get("head_name"):
+            updated[ctx["head_name"]] = ctx["probe"]
+
+    return updated, {}, {}
+
+
+# ── Parallel fusion ───────────────────────────────────────────────────────────
+
+def _parallel_fusion(
+    sub_cfgs, cfg, towers, stage_models, data, split,
+    label_filter, num_classes, device, fold, cfg_stages, _make_loader,
+):
+    nan      = float("nan")
+    bs       = cfg["training"]["batch_size"]
+    bcd_prob = cfg["training"].get("bcd_prob", 0.5)
+
+    # Freeze all prior stage models once, before building any bridges.
+    for m in stage_models.values():
+        for p in m.parameters():
+            p.requires_grad_(False)
+        m.eval()
+
+    # ── Per-sub-stage setup ───────────────────────────────────────────────────
+    contexts = []
+    for sc in sub_cfgs:
+        name       = sc["name"]
+        level      = sc["level"]
+        inputs     = sc["inputs"]
+        epochs     = sc["epochs"]
+        eye_filter = sc.get("eye_filter", None)
+
+        s_train = data.build_shells(split.train, level=level, label_filter=label_filter,
+                                    eye_filter=eye_filter)
+        s_val   = data.build_shells(split.val,   level=level, label_filter=label_filter,
+                                    eye_filter=eye_filter)
+        s_test  = (data.build_shells(split.test, level=level, label_filter=label_filter,
+                                     eye_filter=eye_filter)
+                   if split.test is not None else LoaderShell(entries=[]))
+
+        if not s_val.entries:
+            print(f"  fold{fold+1}: no val samples for stage {name!r}, skipping.", flush=True)
+            continue
+
+        input_dims = resolve_input_dims(inputs, towers, stage_models)
+        bmod       = importlib.import_module(sc["module"])
+        bridge     = getattr(bmod, sc["class"])(input_dims, **sc.get("args", {})).to(device)
+
+        # Head stages for this sub-stage.
+        head_stage_cfgs = [s for s in cfg_stages if s["type"] == "head"
+                           and s.get("train_with") == name]
+        head_models: dict[str, torch.nn.Module] = {}
+        for hs in head_stage_cfgs:
+            h_dim = get_out_dim(hs["input"], towers, {**stage_models, name: bridge})
+            h_mod = importlib.import_module(hs.get("module", "v4.classes.heads.classifier"))
+            h_cls = getattr(h_mod, hs.get("class", "ClassificationHead"))
+            existing = stage_models.get(hs["name"])
+            head_models[hs["name"]] = (
+                existing.to(device) if existing is not None
+                else h_cls(h_dim, num_classes, **hs.get("args", {})).to(device)
+            )
+        for h in head_models.values():
+            for p in h.parameters():
+                p.requires_grad_(True)
+
+        primary_hs_cfg = next((hs for hs in head_stage_cfgs if not hs.get("bcd", False)), None)
+        bcd_head_cfgs  = [hs for hs in head_stage_cfgs if hs.get("bcd", False)]
+
+        if primary_hs_cfg is None:
+            print(f"  WARNING: no primary head for stage {name!r}; skipping.", flush=True)
+            continue
+
+        primary_head = head_models[primary_hs_cfg["name"]]
+
+        train_towers  = sc.get("train_towers", False)
+        direct_inputs = inputs if isinstance(inputs, list) else list(inputs.values())
+        tower_params  = ([p for n in direct_inputs if n in towers
+                          for p in towers[n].parameters()]
+                         if train_towers else [])
+        opt_params    = (tower_params + list(bridge.parameters()) +
+                         [p for h in head_models.values() for p in h.parameters()])
+        opt = torch.optim.Adam(opt_params, lr=cfg["training"]["lr"])
+
+        warmup_cfg = sc.get("warmup", {})
+        wt         = warmup_cfg.get("tower_epochs", 0)
+        wf         = warmup_cfg.get("fused_epochs", 0)
+
+        train_loader = _make_loader(s_train, towers, batch_size=bs, shuffle=True)
+        val_loader   = _make_loader(s_val,   towers, batch_size=bs, shuffle=False)
+        test_loader  = (_make_loader(s_test, towers, batch_size=bs, shuffle=False)
+                        if s_test.entries else None)
+
+        cw = class_weights_from_shell(
+            s_train, num_classes, device,
+            enabled=cfg["training"].get("class_weighted", False),
+        )
+        contexts.append({
+            "name": name, "inputs": inputs, "epochs": epochs,
+            "bridge": bridge, "head_models": head_models,
+            "primary_head": primary_head, "primary_hs_cfg": primary_hs_cfg,
+            "bcd_head_cfgs": bcd_head_cfgs,
+            "opt": opt, "wt": wt, "wf": wf,
+            "train_towers": train_towers, "direct_inputs": direct_inputs,
+            "train_loader": train_loader, "val_loader": val_loader,
+            "test_loader": test_loader, "sc": sc,
+            "class_weights": cw,
+        })
+
+    if not contexts:
+        return stage_models, {}, {}
+
+    max_epochs = max(c["epochs"] for c in contexts)
+
+    # ── Shared epoch loop ─────────────────────────────────────────────────────
+    for epoch in range(max_epochs):
+        for ctx in contexts:
+            if epoch >= ctx["epochs"]:
+                continue
+
+            name   = ctx["name"]
+            bridge = ctx["bridge"]
+            inputs = ctx["inputs"]
+            wt, wf = ctx["wt"], ctx["wf"]
+            phase  = phase_for_epoch(epoch, wt, wf)
+
+            bridge.train()
+            for h in ctx["head_models"].values():
+                h.train()
+            if ctx["train_towers"]:
+                for n in ctx["direct_inputs"]:
+                    if n in towers:
+                        towers[n].train()
+            else:
+                for n in ctx["direct_inputs"]:
+                    if n in towers:
+                        towers[n].eval()
+
+            if hasattr(bridge, "set_phase"):
+                bridge.set_phase(phase)
+
+            total_loss = total_correct = total_n = 0
+
+            for batch in ctx["train_loader"]:
+                y = batch.get("label")
+                if not torch.is_tensor(y):
+                    continue
+                y_t = to_label_tensor(y, device)
+                if y_t.numel() == 0:
+                    continue
+
+                local_embs = {n: towers[n](batch[n].to(device)) for n in inputs
+                              if n in batch and torch.is_tensor(batch[n])}
+                if len(local_embs) != len(inputs):
+                    continue
+                local_embs[name] = bridge(list(local_embs[n] for n in inputs))
+
+                head_logits = {
+                    hs["name"]: ctx["head_models"][hs["name"]](local_embs[hs["input"]])
+                    for hs in ([ctx["primary_hs_cfg"]] + ctx["bcd_head_cfgs"])
+                    if hs["input"] in local_embs
+                }
+
+                if phase == "fused_warmup":
+                    logits = head_logits.get(ctx["primary_hs_cfg"]["name"])
+                elif phase == "tower_warmup" and ctx["bcd_head_cfgs"]:
+                    losses = [F.cross_entropy(head_logits[hs["name"]], y_t,
+                                              weight=ctx["class_weights"])
+                              for hs in ctx["bcd_head_cfgs"] if hs["name"] in head_logits]
+                    if not losses:
+                        continue
+                    loss = sum(losses) / len(losses)
+                    ctx["opt"].zero_grad(); loss.backward(); ctx["opt"].step()
+                    total_loss += loss.item() * len(y_t)
+                    total_n    += len(y_t)
+                    continue
+                else:
+                    if ctx["bcd_head_cfgs"] and _random() < bcd_prob:
+                        logits = head_logits.get(choice(ctx["bcd_head_cfgs"])["name"])
+                    else:
+                        logits = head_logits.get(ctx["primary_hs_cfg"]["name"])
+
+                if logits is None:
+                    continue
+                loss = F.cross_entropy(logits, y_t, weight=ctx["class_weights"])
+                ctx["opt"].zero_grad(); loss.backward(); ctx["opt"].step()
+                total_correct += int((logits.argmax(1) == y_t).sum())
+                total_loss    += loss.item() * len(y_t)
+                total_n       += len(y_t)
+
+            tr_loss = total_loss / total_n if total_n else nan
+            tr_acc  = total_correct / total_n if total_n else nan
+
+            y_v, p_v, _, _ = collect_probs(bridge, ctx["primary_head"], ctx["sc"], towers,
+                                           stage_models, cfg_stages, ctx["val_loader"],
+                                           device, num_classes)
+            _, val_auc, _ = score_arrays(y_v, p_v, num_classes) if y_v.size else (nan, nan, nan)
+            print(
+                f"  fold{fold+1} [{name}] ep{epoch+1:03d}/{ctx['epochs']} [{phase:14s}]"
+                f"  loss={tr_loss:.4f}  acc={tr_acc:.3f}  val_auc={val_auc:.4f}",
+                flush=True,
+            )
+
+    # ── Final eval + collect results ──────────────────────────────────────────
+    updated    = dict(stage_models)
+    all_metrics: dict = {}
+    all_preds:   dict = {}
+
+    for ctx in contexts:
+        name         = ctx["name"]
+        bridge       = ctx["bridge"]
+        primary_head = ctx["primary_head"]
+
+        updated[name] = bridge
+        updated.update(ctx["head_models"])
+
+        y_val, p_val, ids_val, z_val = collect_probs(bridge, primary_head, ctx["sc"], towers,
+                                                      stage_models, cfg_stages, ctx["val_loader"],
+                                                      device, num_classes)
+        val_acc, val_auc, val_n = (score_arrays(y_val, p_val, num_classes)
+                                   if y_val.size else (nan, nan, nan))
+        ext = compute_extended_metrics(y_val, p_val, num_classes) if y_val.size else {}
+
+        val_threshold = 0.5
+        if (cfg["training"].get("tune_binary_threshold")
+                and num_classes == 2 and y_val.size >= 2):
+            val_threshold = tune_binary_threshold(y_val, p_val[:, 1])
+
+        y_te = p_te = ids_te = z_te = None
+        test_auc = test_acc = test_n = nan
+        if ctx["test_loader"] is not None:
+            y_te, p_te, ids_te, z_te = collect_probs(bridge, primary_head, ctx["sc"], towers,
+                                                      stage_models, cfg_stages, ctx["test_loader"],
+                                                      device, num_classes)
+            test_acc, test_auc, test_n = (score_arrays(y_te, p_te, num_classes)
+                                          if y_te.size else (nan, nan, nan))
+
+        all_metrics.update({
+            f"{name}_val_auc":       val_auc,
+            f"{name}_val_acc":       val_acc,
+            f"{name}_val_n":         val_n,
+            f"{name}_val_kappa":     ext.get("kappa", nan),
+            f"{name}_val_mcc":       ext.get("mcc", nan),
+            f"{name}_val_f1":        ext.get("macro_f1", nan),
+            f"{name}_val_threshold": val_threshold,
+            f"{name}_test_auc":      test_auc,
+            f"{name}_test_acc":      test_acc,
+            f"{name}_test_n":        test_n,
+        })
+        all_preds[name] = {
+            "val_y": y_val, "val_p": p_val, "val_ids": ids_val, "val_z": z_val,
+            "test_y": y_te, "test_p": p_te, "test_ids": ids_te, "test_z": z_te,
+        }
+
+    return updated, all_metrics, all_preds

v4/classes/stages/warm.py

@@ -1,11 +1,13 @@
-"""stages/warm — warm stage runner: pre-trains a single tower with a temporary probe."""
+"""stages/warm — warm stage runner: pre-trains a single tower and optional real head."""
 from __future__ import annotations
 
+import importlib
+
 import torch
 import torch.nn.functional as F
 
 from v4.classes.dataset import to_label_tensor
-from v4.classes.stages.helpers import phase_for_epoch
+from v4.classes.stages.helpers import class_weights_from_shell
 
 
 def run(
@@ -20,16 +22,27 @@ def run(
     fold:        int,
     _make_loader,
     _balanced_sampler,
-) -> None:
-    """Pre-train one tower using a temporary linear probe (probe discarded after)."""
+    stage_models: dict | None = None,
+    cfg_stages:   list[dict] | None = None,
+) -> dict:
+    """Pre-train one tower.
+
+    If ``head_name`` is set on the stage config, train that real downstream head
+    and return it in ``stage_models``. Otherwise, fall back to a temporary linear
+    probe for backward-compatible representation warmup.
+    """
     tower_name   = stage_cfg["tower"]
     n_epochs     = stage_cfg.get("epochs", 0)
     level        = stage_cfg["level"]
+    shell_filter = stage_cfg.get("shell_filter", {})
+    stage_models = dict(stage_models or {})
+    cfg_stages   = list(cfg_stages or [])
 
     if n_epochs == 0:
-        return
+        return stage_models
 
-    s_train = data.build_shells(split.train, level=level, label_filter=label_filter)
+    s_train = data.build_shells(split.train, level=level, label_filter=label_filter,
+                                **shell_filter)
     bs      = cfg["training"]["batch_size"]
     loader  = _make_loader(
         s_train, {tower_name: towers[tower_name]},
@@ -41,13 +54,32 @@ def run(
         for p in t.parameters():
             p.requires_grad_(n == tower_name)
 
+    head_name = stage_cfg.get("head_name")
+    if head_name:
+        head_cfg = next((s for s in cfg_stages if s.get("name") == head_name), None)
+        if head_cfg is None:
+            raise ValueError(f"warm stage requested head_name={head_name!r}, but no such head exists")
+        h_mod = importlib.import_module(head_cfg.get("module", "v4.classes.heads.classifier"))
+        h_cls = getattr(h_mod, head_cfg.get("class", "ClassificationHead"))
+        probe = stage_models.get(head_name)
+        if probe is None:
+            probe = h_cls(towers[tower_name].out_dim, num_classes, **head_cfg.get("args", {}))
+        probe = probe.to(device)
+    else:
         probe = torch.nn.Linear(towers[tower_name].out_dim, num_classes).to(device)
+
     opt   = torch.optim.Adam(
         list(towers[tower_name].parameters()) + list(probe.parameters()),
         lr=cfg["training"]["lr"],
     )
 
+    cw = class_weights_from_shell(
+        s_train, num_classes, device,
+        enabled=cfg["training"].get("class_weighted", False),
+    )
+
     towers[tower_name].train()
+    probe.train()
     for epoch in range(n_epochs):
         total_loss = total_correct = total_n = 0
         for batch in loader:
@@ -57,7 +89,7 @@ def run(
                 continue
             y_t    = to_label_tensor(y, device)
             logits = probe(towers[tower_name](x.to(device)))
-            loss   = F.cross_entropy(logits, y_t)
+            loss   = F.cross_entropy(logits, y_t, weight=cw)
             opt.zero_grad(); loss.backward(); opt.step()
             total_loss    += loss.item() * len(y_t)
             total_correct += int((logits.argmax(1) == y_t).sum())
@@ -71,3 +103,7 @@ def run(
     for t in towers.values():
         for p in t.parameters():
             p.requires_grad_(True)
+
+    if head_name:
+        stage_models[head_name] = probe
+    return stage_models

v4/classes/towers/clinical_tower.py

@@ -2,6 +2,25 @@
 
 Self-contained: no v3 dependencies.
 Inherits get_sample dispatch from TowerBase.
+
+Geometry injection (EPC consumption)
+--------------------------------------
+When geom_dim > 0, ClinicalEncoder requests the "geometry_vectors" key from EPC
+during early_pass and appends the geometry features to every clinical vector.
+The input layer is sized to clinical_data.feature_dim + geom_dim automatically.
+
+Config example (cd tower consuming geometry):
+{
+  "name":         "cd",
+  "module":       "v4.classes.towers.clinical_tower",
+  "class":        "ClinicalEncoder",
+  "data_source":  "matrix",
+  "epc_requests": ["geometry_vectors"],
+  "args": {
+    "hidden_dim": 128,
+    "geom_dim":   5
+  }
+}
 """
 from __future__ import annotations
 
@@ -14,7 +33,7 @@ from v4.classes.accessory.se_block import SEBlock
 
 
 class ClinicalEncoder(TowerBase):
-    """MLP over tabular clinical features.
+    """MLP over tabular clinical features, with optional geometry vector injection.
 
     clinical_data : ClinicalDataView — provides feature_dim, vectorize_entity, side_map
     hidden_dim    : output embedding dimensionality
@@ -22,8 +41,12 @@ class ClinicalEncoder(TowerBase):
     use_se        : wrap output with SEBlock channel gating
     se_reduction  : SEBlock bottleneck factor
     se_pre_norm   : apply LayerNorm before SEBlock
+    geom_dim      : number of geometry features to append from EPC (0 = disabled)
+                    requires epc_requests: ["geometry_vectors"] in tower config
     """
 
+    EPC_GEOMETRY_KEY = "geometry_vectors"
+
     def __init__(
         self,
         clinical_data,
@@ -32,11 +55,14 @@ class ClinicalEncoder(TowerBase):
         use_se:      bool  = False,
         se_reduction: int  = 16,
         se_pre_norm: bool  = True,
+        geom_dim:    int   = 0,
     ):
         super().__init__()
         self.clinical_data  = clinical_data
         self._out_dim       = hidden_dim
-        feature_dim        = clinical_data.feature_dim
+        self._geom_dim      = geom_dim
+        self._geom_vectors: dict | None = None  # filled by early_pass when geom_dim > 0
+        feature_dim = clinical_data.feature_dim + geom_dim
 
         self.block0 = nn.Sequential(
             nn.Linear(feature_dim, hidden_dim),
@@ -53,6 +79,12 @@ class ClinicalEncoder(TowerBase):
         self.tower_ln = nn.LayerNorm(hidden_dim) if se_pre_norm else nn.Identity()
         self.tower_se = SEBlock(hidden_dim, reduction=se_reduction, residual=True) if use_se else None
 
+    # ── EPC early_pass ───────────────────────────────────────────────────────
+
+    def early_pass(self, context) -> None:
+        if self._geom_dim > 0:
+            self._geom_vectors = context.require(self.EPC_GEOMETRY_KEY)
+
     # ── TowerBase interface ──────────────────────────────────────────────────
 
     @property
@@ -65,6 +97,14 @@ class ClinicalEncoder(TowerBase):
 
     def _get(self, *ids) -> torch.Tensor:
         arr = self.clinical_data.vectorize_entity(*ids)
+        if self._geom_dim > 0 and self._geom_vectors is not None:
+            pid  = int(ids[0])
+            eye  = str(ids[1]) if len(ids) > 1 else "OD"
+            geom = self._geom_vectors.get(
+                (pid, eye),
+                np.zeros(self._geom_dim, dtype=np.float32),
+            )
+            arr = np.concatenate([arr, geom[: self._geom_dim]])
         return torch.from_numpy(arr.astype(np.float32, copy=False))
 
     # ── nn.Module forward ────────────────────────────────────────────────────

v4/classes/towers/geometry_tower.py

@@ -0,0 +1,237 @@
+"""geometry_tower — GeometrySegEncoder for v4.
+
+A CNN tower that takes a per-eye disc/cup *segmentation map* as input (rather
+than the raw fundus image) and contributes its pooled embedding to fusion.
+
+Seg maps are produced by an underlying loader (GT contour rasterisation or
+UNet inference) during early_pass, then cached per fold.
+
+UNet fine-tuning lives in early_pass too — the loader's `finetune(train_samples)`
+call uses only the training split, then precompute() runs inference on all
+fold samples (train + val + test).
+
+Config example:
+{
+  "name":         "geom",
+  "module":       "v4.classes.towers.geometry_tower",
+  "class":        "GeometrySegEncoder",
+  "data_source":  "image",
+  "args": {
+    "backbone":          "resnet18",
+    "channels":          3,
+    "target_size":       224,
+    "augment":           true,
+    "seg_source":        "gt",
+    "contour_dir":       "Papila/ExpertsSegmentations/Contours"
+  }
+}
+"""
+from __future__ import annotations
+
+import math
+import sys
+from pathlib import Path
+from typing import Any
+
+import numpy as np
+import torch
+from torch import nn
+from torchvision import models
+
+_REPO_ROOT = Path(__file__).resolve().parents[3]
+if str(_REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(_REPO_ROOT))
+
+from v4.classes.accessory.backbones import build_backbone
+from v4.classes.towerbase import TowerBase
+
+
+class GeometrySegEncoder(TowerBase):
+    """CNN tower over disc/cup segmentation maps.
+
+    image_data    : ImageDataView — provides get_image_path(*ids) and side_map.
+                    Must implement build_seg_map_loader(source, **kwargs).
+    backbone      : backbone key (see accessory/backbones.py)
+    channels      : 1 (label map in [0,1]) or 3 (one-hot bg/rim/cup)
+    target_size   : CNN input spatial size (cached arrays already at this size)
+    augment       : random flip + 90° rotation at training time
+    freeze_ratio  : fraction of early backbone blocks to freeze in [0, 1]
+    seg_source    : passed to image_data.build_seg_map_loader (e.g. "gt", "unet")
+    **seg_kwargs  : forwarded to build_seg_map_loader
+    """
+
+    def __init__(
+        self,
+        image_data,
+        backbone:     str   = "resnet18",
+        channels:     int   = 3,
+        target_size:  int   = 224,
+        augment:      bool  = True,
+        freeze_ratio: float = 0.0,
+        seg_source:   str   = "gt",
+        **seg_kwargs: Any,
+    ):
+        super().__init__()
+        self.image_data   = image_data
+        self._channels    = channels
+        self._target_size = target_size
+        self._augment     = augment
+
+        if not hasattr(image_data, "build_seg_map_loader"):
+            raise TypeError(
+                f"GeometrySegEncoder requires image_data to implement "
+                f"build_seg_map_loader(), but {type(image_data).__name__} does not."
+            )
+        loader_kwargs = {
+            "channels":     channels,
+            "target_size":  target_size,
+            **seg_kwargs,
+        }
+        self._loader = image_data.build_seg_map_loader(seg_source, **loader_kwargs)
+        self._seg_cache: dict = {}
+        self._seg_source = seg_source
+
+        self.backbone, self._base_dim, self._blocks = build_backbone(backbone, freeze_ratio)
+        if channels != 3:
+            self._adapt_first_conv(channels)
+
+        print(
+            f"[GeometrySegEncoder] backbone={backbone}  channels={channels}  "
+            f"target_size={target_size}  seg_source={seg_source}",
+            flush=True,
+        )
+
+    # ── TowerBase interface ──────────────────────────────────────────────────
+
+    @property
+    def out_dim(self) -> int:
+        return self._base_dim
+
+    @property
+    def _side_map(self) -> dict[str, str]:
+        return self.image_data.side_map
+
+    def _get(self, *ids) -> torch.Tensor:
+        key = tuple(ids)
+        arr = self._seg_cache.get(key)
+        if arr is None:
+            arr = np.zeros(
+                (self._channels, self._target_size, self._target_size),
+                dtype=np.float32,
+            )
+        if self.training and self._augment:
+            arr = self._augment_array(arr)
+        return torch.from_numpy(np.ascontiguousarray(arr))
+
+    # ── EPC early_pass ───────────────────────────────────────────────────────
+
+    def early_pass(self, context) -> None:
+        data   = context.require("data")
+        split  = context.require("split")
+
+        train_samples = self._collect_samples(split.train, data)
+        all_samples   = self._collect_samples(split.train, data)
+        all_samples  += self._collect_samples(split.val,   data)
+        if split.test is not None:
+            all_samples += self._collect_samples(split.test, data)
+
+        # Reset per-fold state if loader supports it (UNet only).
+        if hasattr(self._loader, "reset_cache"):
+            self._loader.reset_cache()
+        if hasattr(self._loader, "reset_weights"):
+            self._loader.reset_weights()
+        if hasattr(self._loader, "finetune"):
+            self._loader.finetune(train_samples)
+
+        self._loader.precompute(all_samples)
+        self._seg_cache = self._loader.all_seg_maps()
+        print(
+            f"[GeometrySegEncoder] cached {len(self._seg_cache)} seg maps for fold",
+            flush=True,
+        )
+
+    # ── nn.Module forward ────────────────────────────────────────────────────
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        y = self.backbone(x)
+        if y.dim() > 2:
+            y = y.flatten(1)
+        return y
+
+    # ── Utilities ────────────────────────────────────────────────────────────
+
+    def set_freeze_ratio(self, ratio: float) -> None:
+        r        = max(0.0, min(1.0, float(ratio)))
+        n_freeze = int(math.floor(len(self._blocks) * r))
+        for b in self._blocks:
+            for p in b.parameters():
+                p.requires_grad = True
+        for b in self._blocks[:n_freeze]:
+            for p in b.parameters():
+                p.requires_grad = False
+
+    # ── Internals ────────────────────────────────────────────────────────────
+
+    def _collect_samples(self, df, data) -> list:
+        """Build (pid, eye, image_path) tuples from a split DataFrame."""
+        if df is None or len(df) == 0:
+            return []
+        pc = data.patient_col
+        out = []
+        for _, row in df.iterrows():
+            pid = int(row[pc])
+            eye = str(row.get("eyeID", "OD"))
+            out.append((pid, eye, data.image.get_image_path(pid, eye)))
+        return out
+
+    @staticmethod
+    def _augment_array(arr: np.ndarray) -> np.ndarray:
+        """Random flip + 90° rotation on a (C, H, W) seg-map array."""
+        if np.random.rand() < 0.5:
+            arr = arr[:, :, ::-1]
+        if np.random.rand() < 0.5:
+            arr = arr[:, ::-1, :]
+        k = int(np.random.randint(0, 4))
+        if k:
+            arr = np.rot90(arr, k=k, axes=(1, 2))
+        return arr
+
+    def _adapt_first_conv(self, in_channels: int) -> None:
+        """Replace the first Conv2d to accept a non-3-channel input.
+
+        Pretrained weights are averaged across the original input channels and
+        broadcast across the new ones.
+        """
+        first = self._find_first_conv(self.backbone)
+        new   = nn.Conv2d(
+            in_channels,
+            first.out_channels,
+            kernel_size=first.kernel_size,
+            stride=first.stride,
+            padding=first.padding,
+            bias=first.bias is not None,
+        )
+        with torch.no_grad():
+            new.weight.copy_(
+                first.weight.mean(dim=1, keepdim=True).expand_as(new.weight)
+            )
+            if first.bias is not None:
+                new.bias.copy_(first.bias)
+        self._replace_first_conv(self.backbone, new)
+
+    @staticmethod
+    def _find_first_conv(module: nn.Module) -> nn.Conv2d:
+        for m in module.modules():
+            if isinstance(m, nn.Conv2d):
+                return m
+        raise RuntimeError("No Conv2d found in backbone")
+
+    @classmethod
+    def _replace_first_conv(cls, module: nn.Module, new_conv: nn.Conv2d) -> bool:
+        for name, child in module.named_children():
+            if isinstance(child, nn.Conv2d):
+                setattr(module, name, new_conv)
+                return True
+            if cls._replace_first_conv(child, new_conv):
+                return True
+        return False

v4/classes/towers/image_tower.py

@@ -2,30 +2,83 @@
 
 Self-contained: no v3 dependencies.
 Inherits get_sample dispatch from TowerBase.
+
+Geometry injection (EPC supply)
+--------------------------------
+When geometry_source is set, ImageEncoder asks the image_data view for a loader
+via image_data.build_geometry_loader(source, **kwargs).  The view is responsible
+for understanding what that source means for its specific domain (fundus contours,
+U-Net segmentations, cat ear landmarks, etc.).
+
+During early_pass the loader pre-computes all per-entity geometry vectors and
+publishes them to the EarlyPassContext under the key "geometry_vectors"
+({(entity_id...): np.ndarray of length geom_dim}).  ClinicalEncoder (or any
+other tower with epc_requests: ["geometry_vectors"]) can then consume them.
+
+The tower reads feature_dim and feature_names from the loader instance, so it
+can log geometry info without knowing anything about CDR, disc masks, or other
+domain-specific concepts.
+
+Config example:
+{
+  "name":         "img",
+  "module":       "v4.classes.towers.image_tower",
+  "class":        "ImageEncoder",
+  "data_source":  "image",
+  "epc_supplies": ["geometry_vectors"],
+  "args": {
+    "backbone":        "refugelike",
+    "augment":         true,
+    "geometry_source": "gt",
+    "contour_dir":     "Papila/ExpertsSegmentations/Contours"
+  }
+}
 """
 from __future__ import annotations
 
 import math
+import sys
+from pathlib import Path
+from typing import Any
 
 import torch
 from torch import nn
 
+_REPO_ROOT = Path(__file__).resolve().parents[3]
+if str(_REPO_ROOT) not in sys.path:
+    sys.path.insert(0, str(_REPO_ROOT))
+
 from v4.classes.towerbase import TowerBase
 from v4.classes.accessory.backbones import build_backbone
 from v4.classes.accessory.se_block import SEBlock
-from v4.classes.accessory.transforms import build_backbone_transform, build_eval_transform
+from v4.classes.accessory.transforms import (
+    build_backbone_transform, build_eval_transform, build_split_transforms,
+)
 
 
 class ImageEncoder(TowerBase):
     """Vision backbone → pooled feature vector.
 
-    image_data   : ImageDataView — provides load_image(*ids) and side_map
+    image_data        : ImageDataView — provides load_image(*ids) and side_map.
+                        Must implement build_geometry_loader(source, **kwargs)
+                        if geometry_source is set.
     backbone          : backbone key (see accessory/backbones.py)
     freeze_ratio      : fraction of early blocks to freeze in [0, 1]
     use_se            : apply SE attention over the pooled feature vector
     augment           : include random flip/rotation/jitter in the train transform
+    cache_transformed : if True, cache resized + ToTensor'd float32 [0, 1] CHW
+                        tensors per fold.  Per-batch cost drops to augment +
+                        Normalize on tensors only (no PIL, no Resize, no decode).
+                        Memory: ~3 × crop_size² × 4B per cached image.
+                        Cache is rebuilt at the start of every fold via early_pass.
+    geometry_source   : source key passed to image_data.build_geometry_loader()
+                        (e.g. "gt", "unet").  None = geometry disabled.
+    **geom_kwargs     : forwarded verbatim to build_geometry_loader() — e.g.
+                        contour_dir="Papila/ExpertsSegmentations/Contours"
     """
 
+    EPC_GEOMETRY_KEY = "geometry_vectors"
+
     def __init__(
         self,
         image_data,
@@ -35,17 +88,42 @@ class ImageEncoder(TowerBase):
         se_reduction:      int        = 16,
         se_pre_norm:       bool       = True,
         augment:           bool       = True,
+        cache_transformed: bool       = False,
+        geometry_source:   str | None = None,
+        **geom_kwargs: Any,
     ):
         super().__init__()
         self.image_data     = image_data
         self._name          = backbone
         self.backbone, self._base_dim, self._blocks = build_backbone(backbone, freeze_ratio)
+
+        self._cache_transformed = cache_transformed
+        if cache_transformed:
+            self._precache_tf, self._post_train_tf = build_split_transforms(backbone, augment=augment)
+            _,                 self._post_eval_tf  = build_split_transforms(backbone, augment=False)
+            self._tensor_cache: dict[tuple, torch.Tensor] = {}
+        else:
             self.transform      = build_backbone_transform(backbone, augment=augment)
             self.eval_transform = build_eval_transform(backbone)
 
         self.tower_ln = nn.LayerNorm(self._base_dim) if se_pre_norm else nn.Identity()
         self.tower_se = SEBlock(self._base_dim, reduction=se_reduction, residual=True) if use_se else None
 
+        self._geom_loader = None
+        if geometry_source is not None:
+            if not hasattr(image_data, "build_geometry_loader"):
+                raise TypeError(
+                    f"ImageEncoder geometry_source={geometry_source!r} requires "
+                    f"image_data to implement build_geometry_loader(), "
+                    f"but {type(image_data).__name__} does not."
+                )
+            self._geom_loader = image_data.build_geometry_loader(geometry_source, **geom_kwargs)
+            print(
+                f"[ImageEncoder] geometry_source={geometry_source!r}  "
+                f"features={self._geom_loader.feature_names}",
+                flush=True,
+            )
+
     # ── TowerBase interface ──────────────────────────────────────────────────
 
     @property
@@ -57,10 +135,61 @@ class ImageEncoder(TowerBase):
         return self.image_data.side_map
 
     def _get(self, *ids) -> torch.Tensor:
+        if self._cache_transformed:
+            key = tuple(ids)
+            cached = self._tensor_cache.get(key)
+            if cached is None:
+                cached = self._precache_tf(self.image_data.load_image(*ids))
+                self._tensor_cache[key] = cached
+            tail = self._post_train_tf if self.training else self._post_eval_tf
+            return tail(cached)
         img = self.image_data.load_image(*ids)
         t   = self.transform if self.training else self.eval_transform
         return t(img)
 
+    # ── EPC early_pass ───────────────────────────────────────────────────────
+
+    def early_pass(self, context) -> None:
+        """Per-fold setup: warm tensor cache (if enabled), publish geometry vectors."""
+        data = context.require("data")
+
+        if self._cache_transformed:
+            self._tensor_cache.clear()
+            n = self._warm_tensor_cache(data, context.require("split"))
+            print(
+                f"[ImageEncoder] warmed transformed-tensor cache for {n} entries "
+                f"({self._name})",
+                flush=True,
+            )
+
+        if self._geom_loader is None:
+            return
+        self._geom_loader.precompute(data.df, patient_col=data.patient_col)
+        vecs = self._geom_loader.all_vectors()
+        context.put(self.EPC_GEOMETRY_KEY, vecs)
+        print(
+            f"[ImageEncoder] published {len(vecs)} geometry vectors "
+            f"(dim={self._geom_loader.feature_dim}) to EPC key '{self.EPC_GEOMETRY_KEY}'",
+            flush=True,
+        )
+
+    def _warm_tensor_cache(self, data, split) -> int:
+        """Pre-fill the per-tower tensor cache for all entries in this fold's splits."""
+        seen: set[tuple] = set()
+        for df in (split.train, split.val, split.test):
+            if df is None or len(df) == 0:
+                continue
+            pc = data.patient_col
+            for _, row in df.iterrows():
+                pid = int(row[pc])
+                eye = str(row.get("eyeID", "OD"))
+                key = (pid, eye)
+                if key in self._tensor_cache or key in seen:
+                    continue
+                self._tensor_cache[key] = self._precache_tf(self.image_data.load_image(pid, eye))
+                seen.add(key)
+        return len(self._tensor_cache)
+
     # ── nn.Module forward ────────────────────────────────────────────────────
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:

v4/classes/v4_hypertower.py

@@ -29,7 +29,8 @@ sys.path.insert(0, str(REPO_ROOT))
 from v4.classes.dataset import LoaderShell, HTDataset, ht_collate
 from v4.classes.utils import seed_everything, choose_device
 from v4.classes.split_manager import SplitManager
-from v4.classes.stages import warm, fusion
+from v4.classes.stages import warm, fusion, parallel
+from v4.classes.logging.prediction_store import PredictionStore, FeatureStore
 
 
 # ---------------------------------------------------------------------------
@@ -167,28 +168,49 @@ def run_fold(fold: int, splits, cfg: dict, data, num_classes: int, device) -> di
         if hasattr(tower, "early_pass"):
             tower.early_pass(context)
 
+    # Flatten parallel wrappers so sub-stage configs are addressable by name.
+    flat_stages: list[dict] = []
+    for s in cfg_stages:
+        if s.get("type") == "parallel":
+            flat_stages.extend(s["stages"])
+        else:
+            flat_stages.append(s)
+
     stage_models: dict = {}
     fold_result         = {"fold": fold}
+    fold_preds:   dict  = {}   # stage_name → pred_data
 
     for stage_cfg in cfg_stages:
         stype = stage_cfg["type"]
 
         if stype == "warm":
-            warm.run(stage_cfg, towers, data, split, label_filter,
+            stage_models = warm.run(
+                stage_cfg, towers, data, split, label_filter,
                 cfg, num_classes, device, fold,
-                     _make_loader, _balanced_sampler)
+                _make_loader, _balanced_sampler, stage_models, flat_stages,
+            )
 
         elif stype == "fusion":
-            stage_models, metrics = fusion.run(
+            stage_models, metrics, preds = fusion.run(
                 stage_cfg, cfg, towers, stage_models, data, split,
-                label_filter, num_classes, device, fold, cfg_stages,
+                label_filter, num_classes, device, fold, flat_stages,
                 _make_loader,
             )
             fold_result.update(metrics)
+            fold_preds.update(preds)
+
+        elif stype == "parallel":
+            stage_models, metrics, preds = parallel.run(
+                stage_cfg, cfg, towers, stage_models, data, split,
+                label_filter, num_classes, device, fold, flat_stages,
+                _make_loader, _balanced_sampler,
+            )
+            fold_result.update(metrics)
+            fold_preds.update(preds)
 
         # head stages are handled inside fusion.run
 
-    return fold_result
+    return fold_result, fold_preds
 
 
 # ---------------------------------------------------------------------------
@@ -240,7 +262,10 @@ def main():
     out_dir.mkdir(parents=True, exist_ok=True)
 
     eval_stage        = cfg.get("eval_stage", "hb")
+    save_predictions  = cfg.get("save_predictions", False)
+    save_features     = cfg.get("save_features", False)
     fold_results      = []
+    eval_stage_preds  = []   # list[dict] — one per fold, only for eval_stage
     t0 = time.time()
 
     for fold in range(cfg.get("folds", 5)):
@@ -248,8 +273,10 @@ def main():
         n_train = split.train[group_col].nunique() if group_col else len(split.train)
         print(f"\n── fold {fold+1}/{cfg.get('folds', 5)}  train_groups={n_train} ──",
               flush=True)
-        result = run_fold(fold, splits, cfg, data, num_classes, device)
+        result, fold_preds = run_fold(fold, splits, cfg, data, num_classes, device)
         fold_results.append(result)
+        if save_predictions and eval_stage in fold_preds:
+            eval_stage_preds.append(fold_preds[eval_stage])
         print(
             f"  fold{fold+1} DONE"
             f"  val_auc={result.get(f'{eval_stage}_val_auc', float('nan')):.4f}"
@@ -273,6 +300,7 @@ def main():
             "elapsed_s":     round(time.time() - t0, 1),
             "fold_results":  fold_results,
         }
+        out_dir.mkdir(parents=True, exist_ok=True)
         summary_path = out_dir / "summary.json"
         summary_path.write_text(json.dumps(summary, indent=2))
         print(f"\n{'='*60}", flush=True)
@@ -280,6 +308,81 @@ def main():
         print(f"Test AUC: {summary['mean_test_auc']:.4f} ± {summary['std_test_auc']:.4f}", flush=True)
         print(f"Saved: {summary_path}", flush=True)
 
+        if save_predictions and eval_stage_preds:
+            # Collect all unique entity_ids across val+test sets of all folds.
+            seen, all_ids, id_to_y = set(), [], {}
+            for fp in eval_stage_preds:
+                for eid, y in zip(fp["val_ids"], fp["val_y"]):
+                    k = str(eid)
+                    if k not in seen:
+                        seen.add(k); all_ids.append(eid)
+                    id_to_y[k] = int(y)
+                if fp.get("test_ids"):
+                    for eid, y in zip(fp["test_ids"], fp["test_y"]):
+                        k = str(eid)
+                        if k not in seen:
+                            seen.add(k); all_ids.append(eid)
+                        id_to_y[k] = int(y)
+
+            y_true = np.array([id_to_y.get(str(e), -1) for e in all_ids], dtype=np.int64)
+            store  = PredictionStore(n_folds=len(eval_stage_preds), n_classes=num_classes)
+            store.register_phase(
+                phase=eval_stage,
+                entity_ids=all_ids,
+                y_true=y_true,
+                head_names=[f"{eval_stage}_head"],
+                n_epochs=1,
+            )
+            for fold_idx, fp in enumerate(eval_stage_preds):
+                store.record(eval_stage, fold_idx, 0, fp["val_ids"],
+                             f"{eval_stage}_head", fp["val_p"])
+                store.set_split(eval_stage, fold_idx, fp["val_ids"], "val")
+                if fp.get("test_ids"):
+                    store.record(eval_stage, fold_idx, 0, fp["test_ids"],
+                                 f"{eval_stage}_head", fp["test_p"])
+                    store.set_split(eval_stage, fold_idx, fp["test_ids"], "test")
+
+            pred_path = out_dir / "predictions.h5"
+            store.save(pred_path)
+            print(f"Predictions saved: {pred_path}", flush=True)
+
+        if save_features and eval_stage_preds:
+            emb_dim = eval_stage_preds[0]["val_z"].shape[-1]
+            fstore  = FeatureStore(n_folds=len(eval_stage_preds))
+
+            # Build entity_id / y_true universe (same as predictions).
+            seen, all_ids, id_to_y = set(), [], {}
+            for fp in eval_stage_preds:
+                for eid, y in zip(fp["val_ids"], fp["val_y"]):
+                    k = str(eid)
+                    if k not in seen:
+                        seen.add(k); all_ids.append(eid)
+                    id_to_y[k] = int(y)
+                if fp.get("test_ids"):
+                    for eid, y in zip(fp["test_ids"], fp["test_y"]):
+                        k = str(eid)
+                        if k not in seen:
+                            seen.add(k); all_ids.append(eid)
+                        id_to_y[k] = int(y)
+
+            y_true = np.array([id_to_y.get(str(e), -1) for e in all_ids], dtype=np.int64)
+            fstore.register_phase(phase=eval_stage, entity_ids=all_ids, y_true=y_true)
+            fstore.register_head(phase=eval_stage, head=f"{eval_stage}_embedding",
+                                 n_epochs=1, embedding_dim=emb_dim)
+
+            for fold_idx, fp in enumerate(eval_stage_preds):
+                fstore.record(eval_stage, fold_idx, 0, fp["val_ids"],
+                              f"{eval_stage}_embedding", fp["val_z"])
+                fstore.set_split(eval_stage, fold_idx, fp["val_ids"], "val")
+                if fp.get("test_ids") and fp.get("test_z") is not None:
+                    fstore.record(eval_stage, fold_idx, 0, fp["test_ids"],
+                                  f"{eval_stage}_embedding", fp["test_z"])
+                    fstore.set_split(eval_stage, fold_idx, fp["test_ids"], "test")
+
+            feat_path = out_dir / "features.h5"
+            fstore.save(feat_path)
+            print(f"Features saved: {feat_path}", flush=True)
+
 
 if __name__ == "__main__":
     main()

v4/configs/ensemble_fused.json

@@ -1,23 +1,25 @@
 {
   "_notes": [
-    "V4 stage-pipeline: warm → fusion stages with parallel head stages.",
-    "Bridges are pure embedding producers; heads are separate swappable stages.",
-    "BCD-eligible heads are sampled during tower_warmup and main phases.",
-    "eval_stage names the fusion stage whose primary head is used for final metrics."
+    "Shared-weight bilateral architecture: one img tower and one cd tower,",
+    "each called for both eyes. nt trains on all eye-level data.",
+    "hb fuses nt(OD) and nt(OS) at patient level using the same frozen nt."
   ],
-
   "run_name": "v4/ensemble_fused",
   "num_classes": 2,
-  "label_filter":        [0, 1],
+  "label_filter": [
+    0,
+    1
+  ],
   "split_identity_level": 1,
   "eval_stage": "hb",
-  "save_predictions":    false,
+  "save_predictions": true,
   "seed": 1234,
   "folds": 5,
   "fold_seed": 100,
   "output_root": "v4/results",
-  "out_dir_tags":        ["binary", "ntower"],
-
+  "out_dir_tags": [
+    "binary"
+  ],
   "data": {
     "module": "v4.classes.profiles.v4papila",
     "args": {
@@ -26,11 +28,12 @@
       "label_col": "Diagnosis",
       "iop_corr_method": "ratio",
       "iop_drop_raw": true,
-      "exclude_cols":     ["Axial_Length"],
+      "exclude_cols": [
+        "Axial_Length"
+      ],
       "in_memory_cache": true
     }
   },
-
   "towers": [
     {
       "name": "img",
@@ -53,12 +56,12 @@
       }
     }
   ],
-
   "stages": [
     {
       "name": "cd_warm",
       "type": "warm",
       "tower": "cd",
+      "head_name": "cd_aux",
       "level": "eye",
       "epochs": 40
     },
@@ -81,12 +84,20 @@
       "type": "fusion",
       "module": "v4.classes.bridges.fusion_bridge",
       "class": "FusionBridge",
-      "inputs":       ["img", "cd"],
+      "inputs": [
+        "img",
+        "cd"
+      ],
       "level": "eye",
       "epochs": 36,
       "train_towers": true,
-      "warmup":       { "tower_epochs": 3, "fused_epochs": 3 },
-      "args":         { "fusion_dim": 256 }
+      "warmup": {
+        "tower_epochs": 3,
+        "fused_epochs": 3
+      },
+      "args": {
+        "fusion_dim": 256
+      }
     },
     {
       "name": "nt_head",
@@ -99,22 +110,30 @@
       "type": "fusion",
       "module": "v4.classes.bridges.hyperbridge",
       "class": "HyperBridge",
-      "inputs":  { "a": "nt", "b": "nt" },
+      "inputs": {
+        "a": "nt",
+        "b": "nt"
+      },
       "level": "patient",
       "epochs": 10,
-      "args":    { "hidden_dim": 256, "mode": "embedding_mlp" }
+      "args": {
+        "hidden_dim": 256,
+        "mode": "embedding_mlp"
+      }
     },
     {
       "name": "hb_head",
       "type": "head",
       "input": "hb",
-      "train_with": "hb"
+      "train_with": "hb",
+      "args": {
+        "dropout": 0.3
+      }
     }
   ],
-
   "training": {
     "lr": 1e-4,
-    "batch_size":            16,
+    "batch_size": 8,
     "bcd_prob": 0.5,
     "tune_binary_threshold": true
   }

v4/configs/ensemble_fused_geom_gt.json

@@ -0,0 +1,149 @@
+{
+  "_notes": [
+    "Shared-weight bilateral architecture + GT geometry injection.",
+    "img supplies geometry_vectors via EPC; cd consumes them.",
+    "Same shared-weight structure as ensemble_fused.json."
+  ],
+  "run_name": "v4/ensemble_fused_geom_gt",
+  "num_classes": 2,
+  "label_filter": [
+    0,
+    1
+  ],
+  "split_identity_level": 1,
+  "eval_stage": "hb",
+  "save_predictions": true,
+  "seed": 1234,
+  "folds": 5,
+  "fold_seed": 100,
+  "output_root": "v4/results",
+  "out_dir_tags": [
+    "binary"
+  ],
+  "data": {
+    "module": "v4.classes.profiles.v4papila",
+    "args": {
+      "image_dir": "Papila/FundusImages",
+      "clinical_dir": "Papila/ClinicalData",
+      "label_col": "Diagnosis",
+      "iop_corr_method": "ratio",
+      "iop_drop_raw": true,
+      "exclude_cols": [
+        "Axial_Length"
+      ],
+      "in_memory_cache": true
+    }
+  },
+  "towers": [
+    {
+      "name": "img",
+      "module": "v4.classes.towers.image_tower",
+      "class": "ImageEncoder",
+      "data_source": "image",
+      "epc_supplies": [
+        "geometry_vectors"
+      ],
+      "args": {
+        "backbone": "refugelike",
+        "freeze_ratio": 0.0,
+        "augment": true,
+        "geometry_source": "gt",
+        "contour_dir": "Papila/ExpertsSegmentations/Contours"
+      }
+    },
+    {
+      "name": "cd",
+      "module": "v4.classes.towers.clinical_tower",
+      "class": "ClinicalEncoder",
+      "data_source": "matrix",
+      "epc_requests": [
+        "geometry_vectors"
+      ],
+      "args": {
+        "hidden_dim": 128,
+        "geom_dim": 5
+      }
+    }
+  ],
+  "stages": [
+    {
+      "name": "cd_warm",
+      "type": "warm",
+      "tower": "cd",
+      "head_name": "cd_aux",
+      "level": "eye",
+      "epochs": 40
+    },
+    {
+      "name": "img_aux",
+      "type": "head",
+      "input": "img",
+      "train_with": "nt",
+      "bcd": true
+    },
+    {
+      "name": "cd_aux",
+      "type": "head",
+      "input": "cd",
+      "train_with": "nt",
+      "bcd": true
+    },
+    {
+      "name": "nt",
+      "type": "fusion",
+      "module": "v4.classes.bridges.fusion_bridge",
+      "class": "FusionBridge",
+      "inputs": [
+        "img",
+        "cd"
+      ],
+      "level": "eye",
+      "epochs": 36,
+      "train_towers": true,
+      "warmup": {
+        "tower_epochs": 3,
+        "fused_epochs": 3
+      },
+      "args": {
+        "fusion_dim": 256
+      }
+    },
+    {
+      "name": "nt_head",
+      "type": "head",
+      "input": "nt",
+      "train_with": "nt"
+    },
+    {
+      "name": "hb",
+      "type": "fusion",
+      "module": "v4.classes.bridges.hyperbridge",
+      "class": "HyperBridge",
+      "inputs": {
+        "a": "nt",
+        "b": "nt"
+      },
+      "level": "patient",
+      "epochs": 10,
+      "args": {
+        "hidden_dim": 256,
+        "mode": "embedding_mlp"
+      }
+    },
+    {
+      "name": "hb_head",
+      "type": "head",
+      "input": "hb",
+      "train_with": "hb",
+      "args": {
+        "dropout": 0.3
+      }
+    }
+  ],
+  "training": {
+    "lr": 1e-4,
+    "batch_size": 8,
+    "bcd_prob": 0.5,
+    "tune_binary_threshold": true
+  }
+}

v4/configs/geometry_solo.json

@@ -0,0 +1,85 @@
+{
+  "_notes": [
+    "Standalone geometry tower: GeometrySegEncoder over UNet-derived seg maps,",
+    "MonoBridge passthrough, classification head.  Same architecture as the",
+    "seg_cnn_unet_ft_mono test run, used as a 10-rep baseline for comparison",
+    "against tritower configurations."
+  ],
+  "run_name": "v4/geometry_solo",
+  "num_classes": 2,
+  "label_filter": [0, 1],
+  "split_identity_level": 1,
+  "eval_stage": "geom_fuse",
+  "save_predictions": true,
+  "seed": 1234,
+  "folds": 5,
+  "fold_seed": 100,
+  "output_root": "v4/results",
+  "out_dir_tags": ["binary"],
+
+  "data": {
+    "module": "v4.classes.profiles.v4papila",
+    "args": {
+      "image_dir": "Papila/FundusImages",
+      "clinical_dir": "Papila/ClinicalData",
+      "label_col": "Diagnosis",
+      "iop_corr_method": "ratio",
+      "iop_drop_raw": true,
+      "exclude_cols": ["Axial_Length"],
+      "in_memory_cache": false
+    }
+  },
+
+  "towers": [
+    {
+      "name": "geom",
+      "module": "v4.classes.towers.geometry_tower",
+      "class": "GeometrySegEncoder",
+      "data_source": "image",
+      "args": {
+        "backbone":            "resnet18",
+        "channels":            3,
+        "target_size":         224,
+        "augment":             true,
+        "freeze_ratio":        0.0,
+        "seg_source":          "unet",
+        "weights_path":        "models/v2/refuge/segmentation/per_image/best.pt",
+        "contour_dir":         "Papila/ExpertsSegmentations/Contours",
+        "unet_size":           512,
+        "normalize":           "per_image",
+        "threshold":           0.5,
+        "crop_to_disc":        true,
+        "finetune_epochs":     10,
+        "finetune_lr":         1e-5,
+        "finetune_batch_size": 4
+      }
+    }
+  ],
+
+  "stages": [
+    {
+      "name": "geom_fuse",
+      "type": "fusion",
+      "module": "v4.classes.bridges.mono_bridge",
+      "class": "MonoBridge",
+      "inputs": ["geom"],
+      "level": "eye",
+      "epochs": 60,
+      "train_towers": true,
+      "args": { "use_ln": false }
+    },
+    {
+      "name": "geom_head",
+      "type": "head",
+      "input": "geom_fuse",
+      "train_with": "geom_fuse",
+      "args": { "dropout": 0.3 }
+    }
+  ],
+
+  "training": {
+    "lr": 1e-4,
+    "batch_size": 16,
+    "tune_binary_threshold": true
+  }
+}

v4/configs/tritower.json

@@ -0,0 +1,163 @@
+{
+  "_notes": [
+    "Tri-tower shared-weight bilateral: img + cd + geom (UNet seg maps).",
+    "Default parameters; serves as both a 10-rep baseline and as the base",
+    "config for the bcd/cw/nt-epochs grid search.",
+    "geom is a GeometrySegEncoder over per-fold fine-tuned UNet predictions.",
+    "geom trains alongside img inside nt with the standard tower_warmup phase",
+    "(no separate geom_warm stage).  hb fuses nt(OD) and nt(OS) at patient level."
+  ],
+  "run_name": "v4/tritower",
+  "num_classes": 2,
+  "label_filter": [0, 1],
+  "split_identity_level": 1,
+  "eval_stage": "hb",
+  "save_predictions": true,
+  "seed": 1234,
+  "folds": 5,
+  "fold_seed": 100,
+  "output_root": "v4/results",
+  "out_dir_tags": ["binary"],
+
+  "data": {
+    "module": "v4.classes.profiles.v4papila",
+    "args": {
+      "image_dir": "Papila/FundusImages",
+      "clinical_dir": "Papila/ClinicalData",
+      "label_col": "Diagnosis",
+      "iop_corr_method": "ratio",
+      "iop_drop_raw": true,
+      "exclude_cols": ["Axial_Length"],
+      "in_memory_cache": true
+    }
+  },
+
+  "towers": [
+    {
+      "name": "img",
+      "module": "v4.classes.towers.image_tower",
+      "class": "ImageEncoder",
+      "data_source": "image",
+      "args": {
+        "backbone": "refugelike",
+        "freeze_ratio": 0.0,
+        "augment": true
+      }
+    },
+    {
+      "name": "cd",
+      "module": "v4.classes.towers.clinical_tower",
+      "class": "ClinicalEncoder",
+      "data_source": "matrix",
+      "args": {
+        "hidden_dim": 128
+      }
+    },
+    {
+      "name": "geom",
+      "module": "v4.classes.towers.geometry_tower",
+      "class": "GeometrySegEncoder",
+      "data_source": "image",
+      "args": {
+        "backbone":            "resnet18",
+        "channels":            3,
+        "target_size":         224,
+        "augment":             true,
+        "freeze_ratio":        0.0,
+        "seg_source":          "unet",
+        "weights_path":        "models/v2/refuge/segmentation/per_image/best.pt",
+        "contour_dir":         "Papila/ExpertsSegmentations/Contours",
+        "unet_size":           512,
+        "normalize":           "per_image",
+        "threshold":           0.5,
+        "crop_to_disc":        true,
+        "finetune_epochs":     10,
+        "finetune_lr":         1e-5,
+        "finetune_batch_size": 4
+      }
+    }
+  ],
+
+  "stages": [
+    {
+      "name": "cd_warm",
+      "type": "warm",
+      "tower": "cd",
+      "head_name": "cd_aux",
+      "level": "eye",
+      "epochs": 40
+    },
+    {
+      "name": "img_aux",
+      "type": "head",
+      "input": "img",
+      "train_with": "nt",
+      "bcd": true
+    },
+    {
+      "name": "cd_aux",
+      "type": "head",
+      "input": "cd",
+      "train_with": "nt",
+      "bcd": true
+    },
+    {
+      "name": "geom_aux",
+      "type": "head",
+      "input": "geom",
+      "train_with": "nt",
+      "bcd": true
+    },
+    {
+      "name": "nt",
+      "type": "fusion",
+      "module": "v4.classes.bridges.fusion_bridge",
+      "class": "FusionBridge",
+      "inputs": ["img", "cd", "geom"],
+      "level": "eye",
+      "epochs": 36,
+      "train_towers": true,
+      "warmup": {
+        "tower_epochs": 3,
+        "fused_epochs": 3
+      },
+      "args": {
+        "fusion_dim": 256
+      }
+    },
+    {
+      "name": "nt_head",
+      "type": "head",
+      "input": "nt",
+      "train_with": "nt"
+    },
+    {
+      "name": "hb",
+      "type": "fusion",
+      "module": "v4.classes.bridges.hyperbridge",
+      "class": "HyperBridge",
+      "inputs": { "a": "nt", "b": "nt" },
+      "level": "patient",
+      "epochs": 10,
+      "args": {
+        "hidden_dim": 256,
+        "mode": "embedding_mlp"
+      }
+    },
+    {
+      "name": "hb_head",
+      "type": "head",
+      "input": "hb",
+      "train_with": "hb",
+      "args": { "dropout": 0.3 }
+    }
+  ],
+
+  "training": {
+    "lr": 1e-4,
+    "batch_size": 8,
+    "bcd_prob": 0.5,
+    "tune_binary_threshold": true,
+    "class_weighted": false
+  }
+}

v4/distributed/batch_dispatch.py

@@ -0,0 +1,287 @@
+"""
+v4 batch dispatch — merge a batch.json with a base config and submit jobs.
+
+Each batch entry specifies only what differs from the base config. Each entry
+is dispatched as N_REPS independent jobs (full 5-fold CV per rep, different seeds),
+writing to {run_name}/rep00/, rep01/, etc.
+
+Usage:
+    python -m v4.distributed.batch_dispatch \\
+        --server http://apollo:8765 \\
+        --token <secret> \\
+        --config v4/configs/ensemble_fused.json \\
+        --batch v4/scripts/experiments/my_batch.json \\
+        [--reps 10] \\
+        [--seed-start 1234] \\
+        [--seed-step 100] \\
+        [--fold-seed-start 100] \\
+        [--fold-seed-step 100] \\
+        [--output-root v4/results] \\
+        [--priority 0] \\
+        [--dry-run]
+
+batch.json format:
+    [
+      {
+        "run_name":  "experiments/lr_sweep/lr1e3",   // required
+        "overrides": {                               // optional — deep-merged into base
+          "training": { "lr": 0.001 }
+        },
+        "stage_overrides": {                         // optional — patched by stage name
+          "nt": { "epochs": 40 }
+        },
+        "reps":     10,                              // optional — overrides --reps
+        "priority": 0                                // optional
+      }
+    ]
+
+Global flags (can also be set via env vars):
+    --server  HT_SERVER
+    --token   HT_TOKEN
+"""
+from __future__ import annotations
+
+import argparse
+import copy
+import json
+import os
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+import requests
+
+# ──────────────────────────────────────────────────────────────
+# Config helpers
+# ──────────────────────────────────────────────────────────────
+
+def deep_merge(base: dict, override: dict) -> dict:
+    """Recursively merge override into a copy of base.
+
+    - Dicts are merged recursively.
+    - All other types (scalars, lists) are replaced by the override value.
+    """
+    result = copy.deepcopy(base)
+    for k, v in override.items():
+        if k in result and isinstance(result[k], dict) and isinstance(v, dict):
+            result[k] = deep_merge(result[k], v)
+        else:
+            result[k] = copy.deepcopy(v)
+    return result
+
+
+def apply_stage_overrides(stages: list[dict], stage_overrides: dict) -> list[dict]:
+    """Patch individual stages by name without replacing the entire list."""
+    stages = copy.deepcopy(stages)
+    for stage in stages:
+        name = stage.get("name")
+        if name in stage_overrides:
+            merged = deep_merge(stage, stage_overrides[name])
+            stage.clear()
+            stage.update(merged)
+    return stages
+
+
+def build_config(base_cfg: dict, entry: dict, rep: int, seed: int, fold_seed: int,
+                 output_root: str) -> dict:
+    """Produce the final merged config for one rep of one batch entry."""
+    cfg = copy.deepcopy(base_cfg)
+
+    # Deep-merge top-level overrides
+    cfg = deep_merge(cfg, entry.get("overrides", {}))
+
+    # Patch individual stages by name
+    if "stage_overrides" in entry and "stages" in cfg:
+        cfg["stages"] = apply_stage_overrides(cfg["stages"], entry["stage_overrides"])
+
+    # Stamp run_name, model seed, split seed, output_root.
+    base_run_name = entry["run_name"]
+    cfg["run_name"]    = f"{base_run_name}/rep{rep:02d}"
+    cfg["seed"]        = seed
+    cfg["fold_seed"]   = fold_seed
+    cfg["output_root"] = output_root
+
+    return cfg
+
+
+def save_dispatched_config(cfg: dict, base_run_name: str, rep: int,
+                           repo_root: Path) -> Path:
+    """Write the merged config to v4/configs/dispatched/ and return its path."""
+    out_dir = repo_root / "v4" / "configs" / "dispatched" / base_run_name
+    out_dir.mkdir(parents=True, exist_ok=True)
+    path = out_dir / f"rep{rep:02d}.json"
+    path.write_text(json.dumps(cfg, indent=2))
+    return path
+
+
+# ──────────────────────────────────────────────────────────────
+# Server HTTP helper
+# ──────────────────────────────────────────────────────────────
+
+class _API:
+    def __init__(self, base_url: str, token: str):
+        self.base_url = base_url.rstrip("/")
+        self._h = {"x-token": token}
+
+    def post(self, path: str, body: dict) -> dict:
+        r = requests.post(f"{self.base_url}{path}", headers=self._h,
+                          json=body, timeout=10)
+        r.raise_for_status()
+        return r.json()
+
+
+# ──────────────────────────────────────────────────────────────
+# Dispatch
+# ──────────────────────────────────────────────────────────────
+
+def dispatch_batch(
+    api: _API,
+    base_cfg: dict,
+    batch: list[dict],
+    *,
+    default_reps:    int,
+    seed_start:      int,
+    seed_step:       int,
+    fold_seed_start: int,
+    fold_seed_step:  int,
+    output_root:     str,
+    default_priority: int,
+    server_path:     str,
+    repo_root:       Path,
+    dry_run:         bool,
+) -> None:
+    total = skipped = 0
+    for entry in batch:
+        run_name = entry["run_name"]
+        reps     = entry.get("reps", default_reps)
+        priority = entry.get("priority", default_priority)
+
+        print(f"\n[dispatch] {run_name}  ({reps} reps)")
+
+        for rep in range(reps):
+            seed      = seed_start + rep * seed_step
+            fold_seed = fold_seed_start + rep * fold_seed_step
+            cfg       = build_config(base_cfg, entry, rep, seed, fold_seed, output_root)
+            cfg_path = save_dispatched_config(cfg, run_name, rep, repo_root)
+
+            rel_path        = cfg_path.relative_to(repo_root)
+            server_cfg_path = str(rel_path)
+
+            job_body = {
+                "run_name":   run_name,
+                "module":     "v4.classes.v4_hypertower",
+                "args":       ["--config", server_cfg_path],
+                "output_dir": output_root,
+                "priority":   priority,
+            }
+
+            if dry_run:
+                print(
+                    f"  [dry-run] rep{rep:02d}  seed={seed}  "
+                    f"fold_seed={fold_seed}  config={server_cfg_path}"
+                )
+                total += 1
+            else:
+                resp = api.post("/jobs", job_body)
+                if resp.get("skipped"):
+                    print(f"  rep{rep:02d}  [skip — results exist on disk]")
+                    skipped += 1
+                elif resp.get("duplicate"):
+                    print(f"  rep{rep:02d}  [skip — already queued]  job_id={resp['job_id']}")
+                    skipped += 1
+                else:
+                    print(
+                        f"  rep{rep:02d}  seed={seed}  fold_seed={fold_seed}  "
+                        f"job_id={resp['job_id']}  config={server_cfg_path}"
+                    )
+                    total += 1
+
+    action = "would submit" if dry_run else "submitted"
+    skip_note = f"  ({skipped} already queued/done, skipped)" if skipped else ""
+    print(f"\n[dispatch] {action} {total} jobs across {len(batch)} experiment(s){skip_note}")
+
+
+# ──────────────────────────────────────────────────────────────
+# Entry point
+# ──────────────────────────────────────────────────────────────
+
+def main():
+    repo_root = Path(__file__).resolve().parents[2]
+
+    ap = argparse.ArgumentParser(
+        description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter
+    )
+    ap.add_argument("--server", default=os.environ.get("HT_SERVER", ""),
+                    help="Server URL (or set HT_SERVER)")
+    ap.add_argument("--token", default=os.environ.get("HT_TOKEN", ""),
+                    help="Shared secret (or set HT_TOKEN)")
+    ap.add_argument("--config", required=True,
+                    help="Base config JSON file")
+    ap.add_argument("--batch", required=True,
+                    help="Batch JSON file listing experiments")
+    ap.add_argument("--reps", type=int, default=10,
+                    help="Repetitions per experiment (default: 10)")
+    ap.add_argument("--seed-start", type=int, default=1234,
+                    help="Seed for rep00 (default: 1234)")
+    ap.add_argument("--seed-step", type=int, default=100,
+                    help="Seed increment per rep (default: 100)")
+    ap.add_argument("--fold-seed-start", type=int, default=100,
+                    help="Split fold_seed for rep00 (default: 100; matches v3)")
+    ap.add_argument("--fold-seed-step", type=int, default=100,
+                    help="Split fold_seed increment per rep (default: 100; matches v3)")
+    ap.add_argument("--output-root", default="v4/results",
+                    help="Output root written into each config (default: v4/results)")
+    ap.add_argument("--server-path", default="",
+                    help="Absolute path to hypertower root on server "
+                         "(used to build config paths in job args; "
+                         "if omitted, relative paths are used)")
+    ap.add_argument("--priority", type=int, default=0,
+                    help="Default job priority (default: 0)")
+    ap.add_argument("--dry-run", action="store_true",
+                    help="Print jobs without submitting")
+    args = ap.parse_args()
+
+    if not args.dry_run:
+        if not args.server:
+            ap.error("--server is required (or set HT_SERVER)")
+        if not args.token:
+            ap.error("--token is required (or set HT_TOKEN)")
+
+    cfg_path = Path(args.config)
+    if not cfg_path.is_absolute():
+        cfg_path = repo_root / cfg_path
+    base_cfg = json.loads(cfg_path.read_text())
+
+    batch_path = Path(args.batch)
+    if not batch_path.is_absolute():
+        batch_path = repo_root / batch_path
+    batch = json.loads(batch_path.read_text())
+
+    if not isinstance(batch, list):
+        sys.exit("batch.json must be a JSON array")
+    for i, entry in enumerate(batch):
+        if "run_name" not in entry:
+            sys.exit(f"batch entry {i} is missing required 'run_name'")
+
+    api = _API(args.server, args.token) if not args.dry_run else None
+
+    dispatch_batch(
+        api,
+        base_cfg,
+        batch,
+        default_reps=args.reps,
+        seed_start=args.seed_start,
+        seed_step=args.seed_step,
+        fold_seed_start=args.fold_seed_start,
+        fold_seed_step=args.fold_seed_step,
+        output_root=args.output_root,
+        default_priority=args.priority,
+        server_path=args.server_path,
+        repo_root=repo_root,
+        dry_run=args.dry_run,
+    )
+
+
+if __name__ == "__main__":
+    main()

v4/distributed/cli.py

@@ -0,0 +1,347 @@
+"""
+HyperTower distributed job CLI — submit jobs, view status.
+
+Usage:
+    # View all connected clients
+    python -m v4.distributed.cli clients
+
+    # View a specific client
+    python -m v4.distributed.cli clients <client_id>
+
+    # Live monitoring
+    python -m v4.distributed.cli clients --watch
+
+    # View jobs (optionally filter by state)
+    python -m v4.distributed.cli jobs [--state pending|running|done|failed]
+
+    # Submit a job
+    python -m v4.distributed.cli submit \\
+        --run-name v4/ensemble_fused \\
+        -- --config v4/configs/ensemble_fused.json
+
+    # Submit all jobs from a batch file (JSON)
+    python -m v4.distributed.cli submit-batch jobs.json
+
+    # Cancel a pending job
+    python -m v4.distributed.cli cancel <job_id>
+
+Global flags (can also be set via env vars):
+    --server  HT_SERVER   e.g. http://apollo:8765
+    --token   HT_TOKEN
+"""
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sys
+import time
+from datetime import datetime, timezone
+from typing import Optional
+
+import requests
+
+# ──────────────────────────────────────────────────────────────
+# HTTP helpers
+# ──────────────────────────────────────────────────────────────
+
+class _API:
+    def __init__(self, base_url: str, token: str):
+        self.base_url = base_url.rstrip("/")
+        self._h = {"x-token": token}
+
+    def get(self, path: str, **params) -> object:
+        r = requests.get(f"{self.base_url}{path}", headers=self._h,
+                         params=params, timeout=10)
+        r.raise_for_status()
+        return r.json()
+
+    def post(self, path: str, body: dict) -> object:
+        r = requests.post(f"{self.base_url}{path}", headers=self._h,
+                          json=body, timeout=10)
+        r.raise_for_status()
+        return r.json()
+
+    def delete(self, path: str) -> object:
+        r = requests.delete(f"{self.base_url}{path}", headers=self._h, timeout=10)
+        r.raise_for_status()
+        return r.json()
+
+
+# ──────────────────────────────────────────────────────────────
+# Formatting helpers
+# ──────────────────────────────────────────────────────────────
+
+def _ago(ts: Optional[str]) -> str:
+    if not ts:
+        return "-"
+    try:
+        dt = datetime.fromisoformat(ts)
+        delta = datetime.now(timezone.utc) - dt
+        secs = int(delta.total_seconds())
+        if secs < 60:
+            return f"{secs}s ago"
+        elif secs < 3600:
+            return f"{secs//60}m ago"
+        else:
+            return f"{secs//3600}h{(secs%3600)//60}m ago"
+    except Exception:
+        return ts
+
+
+def _table(rows: list[list[str]], headers: list[str]):
+    widths = [max(len(str(r[i])) for r in ([headers] + rows)) for i in range(len(headers))]
+    sep = "  "
+    def _row(r):
+        return sep.join(str(r[i]).ljust(widths[i]) for i in range(len(r)))
+    print(_row(headers))
+    print("-" * (sum(widths) + len(sep) * (len(widths) - 1)))
+    for r in rows:
+        print(_row(r))
+
+
+# ──────────────────────────────────────────────────────────────
+# Subcommands
+# ──────────────────────────────────────────────────────────────
+
+def _clients_table(api: _API) -> str:
+    clients = api.get("/clients")
+    if not clients:
+        return "No clients connected."
+    rows = []
+    for c in clients:
+        s = c["status"]
+        parts = []
+        if s.get("fold") is not None:
+            parts.append(f"fold{s['fold']}")
+        if s.get("stage"):
+            parts.append(s["stage"])
+        if s.get("epoch") is not None:
+            parts.append(f"ep{s['epoch']}/{s.get('total_epochs', '?')}")
+        if s.get("last_val_auc") is not None:
+            parts.append(f"auc={s['last_val_auc']:.4f}")
+        prog = " ".join(parts) if parts else "-"
+        rows.append([
+            c["client_id"],
+            c["hostname"],
+            c["gpu_info"][:30],
+            s["state"],
+            s.get("run_name") or "-",
+            prog,
+            _ago(c["last_seen"]),
+        ])
+    headers = ["ID", "HOST", "GPU", "STATE", "RUN", "PROGRESS", "SEEN"]
+    widths = [max(len(str(r[i])) for r in ([headers] + rows)) for i in range(len(headers))]
+    sep = "  "
+    lines = []
+    lines.append(sep.join(str(h).ljust(widths[i]) for i, h in enumerate(headers)))
+    lines.append("-" * (sum(widths) + len(sep) * (len(widths) - 1)))
+    for r in rows:
+        lines.append(sep.join(str(r[i]).ljust(widths[i]) for i in range(len(r))))
+    return "\n".join(lines)
+
+
+def cmd_clients(api: _API, args):
+    if hasattr(args, "client_id") and args.client_id:
+        data = api.get(f"/clients/{args.client_id}")
+        s = data["status"]
+        print(f"client_id  : {data['client_id']}")
+        print(f"hostname   : {data['hostname']}")
+        print(f"gpu        : {data['gpu_info']}")
+        print(f"last_seen  : {_ago(data['last_seen'])}")
+        print(f"state      : {s['state']}")
+        if s.get("job_id"):
+            print(f"job        : {s['job_id']}  ({s.get('run_name', '')})")
+        if s.get("fold") is not None:
+            print(f"progress   : fold {s['fold']}  stage {s.get('stage', '?')}  "
+                  f"ep {s.get('epoch', '?')}/{s.get('total_epochs', '?')}  "
+                  f"val_auc={s.get('last_val_auc', '?')}")
+        return
+
+    watch = getattr(args, "watch", False)
+    interval = getattr(args, "interval", 5)
+
+    if not watch:
+        print(_clients_table(api))
+        return
+
+    try:
+        while True:
+            now = datetime.now().strftime("%H:%M:%S")
+            print(f"\033[H\033[2J", end="")
+            print(f"HyperTower clients  [{now}]  (Ctrl-C to exit)\n")
+            print(_clients_table(api))
+            time.sleep(interval)
+    except KeyboardInterrupt:
+        print("\nStopped.")
+
+
+def cmd_jobs(api: _API, args):
+    params = {}
+    if hasattr(args, "state") and args.state:
+        params["state"] = args.state
+    jobs = api.get("/jobs", **params)
+    if not jobs:
+        print("No jobs.")
+        return
+
+    try:
+        clients = api.get("/clients")
+        id_to_client = {c["client_id"]: c for c in clients}
+    except Exception:
+        id_to_client = {}
+
+    rows = []
+    for j in jobs:
+        attempts  = j.get("attempts", 0)
+        client_id = j.get("assigned_to")
+        client    = id_to_client.get(client_id) if client_id else None
+
+        client_label = (client["hostname"] if client else client_id) if client_id else "-"
+
+        progress = "-"
+        if client:
+            s     = client.get("status", {})
+            parts = []
+            if s.get("fold") is not None:
+                parts.append(f"fold{s['fold']}")
+            if s.get("stage"):
+                parts.append(s["stage"])
+            if s.get("epoch") is not None:
+                parts.append(f"ep{s['epoch']}/{s.get('total_epochs', '?')}")
+            if parts:
+                progress = " ".join(parts)
+
+        rows.append([
+            j["job_id"][:12],
+            j["run_name"],
+            j["state"],
+            f"{attempts}" if attempts else "-",
+            client_label,
+            progress,
+            _ago(j["created_at"]),
+            _ago(j.get("started_at")),
+            _ago(j.get("completed_at")),
+        ])
+    _table(rows, ["JOB_ID", "RUN_NAME", "STATE", "TRIES", "CLIENT", "PROGRESS", "CREATED", "STARTED", "DONE"])
+    pending = sum(1 for j in jobs if j["state"] == "pending")
+    running = sum(1 for j in jobs if j["state"] == "running")
+    done    = sum(1 for j in jobs if j["state"] == "done")
+    failed  = sum(1 for j in jobs if j["state"] == "failed")
+    print(f"\n  {len(jobs)} total  |  {pending} pending  {running} running  {done} done  {failed} failed")
+
+
+def cmd_submit(api: _API, args):
+    body = {
+        "run_name": args.run_name,
+        "module": args.module,
+        "args": args.run_args,
+        "output_dir": args.output_dir,
+        "priority": args.priority,
+    }
+    resp = api.post("/jobs", body)
+    print(f"Queued job {resp['job_id']} ({args.run_name})")
+
+
+def cmd_submit_batch(api: _API, args):
+    with open(args.batch_file) as f:
+        jobs = json.load(f)
+    for job in jobs:
+        resp = api.post("/jobs", job)
+        print(f"Queued {resp['job_id']}  ({job['run_name']})")
+
+
+def cmd_cancel(api: _API, args):
+    resp = api.delete(f"/jobs/{args.job_id}")
+    print(f"Cancelled {args.job_id}" if resp.get("ok") else resp)
+
+
+def cmd_clear(api: _API, args):
+    body: dict = {}
+    if args.all:
+        body["all"] = True
+    elif args.run_name:
+        body["run_name"] = args.run_name
+    else:
+        body["states"] = args.states or ["done", "failed", "cancelled"]
+    resp = api.post("/jobs/clear", body)
+    print(f"Cleared {resp['cleared']} jobs.")
+
+
+# ──────────────────────────────────────────────────────────────
+# Parser
+# ──────────────────────────────────────────────────────────────
+
+def main():
+    ap = argparse.ArgumentParser(description=__doc__,
+                                  formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--server", default=os.environ.get("HT_SERVER", ""),
+                    help="Server URL (or set HT_SERVER)")
+    ap.add_argument("--token", default=os.environ.get("HT_TOKEN", ""),
+                    help="Shared secret (or set HT_TOKEN)")
+
+    sub = ap.add_subparsers(dest="cmd", required=True)
+
+    # clients
+    p_cl = sub.add_parser("clients", help="List clients or inspect one")
+    p_cl.add_argument("client_id", nargs="?")
+    p_cl.add_argument("--watch", "-w", action="store_true",
+                      help="Live monitoring mode — refresh every --interval seconds")
+    p_cl.add_argument("--interval", "-n", type=int, default=5,
+                      help="Refresh interval in seconds for --watch (default: 5)")
+
+    # jobs
+    p_j = sub.add_parser("jobs", help="List jobs")
+    p_j.add_argument("--state", choices=["pending", "running", "done", "failed", "cancelled"])
+
+    # submit
+    p_s = sub.add_parser("submit", help="Submit a single job")
+    p_s.add_argument("--run-name", required=True)
+    p_s.add_argument("--module", default="v4.classes.v4_hypertower")
+    p_s.add_argument("--output-dir", default="v4/results")
+    p_s.add_argument("--priority", type=int, default=0)
+    p_s.add_argument("run_args", nargs=argparse.REMAINDER,
+                     help="Args after '--' are forwarded to the module")
+
+    # submit-batch
+    p_b = sub.add_parser("submit-batch", help="Submit jobs from a JSON file")
+    p_b.add_argument("batch_file")
+
+    # cancel
+    p_c = sub.add_parser("cancel", help="Cancel a pending job")
+    p_c.add_argument("job_id")
+
+    # clear
+    p_cl2 = sub.add_parser("clear", help="Delete jobs by run-name, state, or everything")
+    p_cl2.add_argument("--run-name", default=None, help="Delete all jobs with this run-name")
+    p_cl2.add_argument("--states", nargs="+",
+                       default=None,
+                       choices=["done", "failed", "cancelled", "pending", "running"],
+                       help="Delete jobs in these states (default: done+failed+cancelled)")
+    p_cl2.add_argument("--all", action="store_true", help="Delete ALL jobs")
+
+    args = ap.parse_args()
+
+    if not args.server:
+        ap.error("--server is required (or set HT_SERVER)")
+    if not args.token:
+        ap.error("--token is required (or set HT_TOKEN)")
+
+    if hasattr(args, "run_args") and args.run_args and args.run_args[0] == "--":
+        args.run_args = args.run_args[1:]
+
+    api = _API(args.server, args.token)
+
+    dispatch = {
+        "clients": cmd_clients,
+        "jobs": cmd_jobs,
+        "submit": cmd_submit,
+        "submit-batch": cmd_submit_batch,
+        "cancel": cmd_cancel,
+        "clear": cmd_clear,
+    }
+    dispatch[args.cmd](api, args)
+
+
+if __name__ == "__main__":
+    main()

v4/distributed/client.py

@@ -0,0 +1,407 @@
+"""
+HyperTower distributed job client daemon.
+
+Registers with the server, polls for jobs, syncs code, runs training,
+uploads results, and loops. Parses stdout to stream live status.
+
+Usage:
+    python -m v4.distributed.client \\
+        --server http://apollo:8765 \\
+        --token <secret> \\
+        --server-ssh rpotter@apollo \\
+        --server-path /home/rpotter/hypertower \\
+        [--local-path ~/hypertower] \\
+        [--poll-interval 15]
+
+Compatibility test (verify GPU env, 1-fold dry-run):
+    python -m v4.distributed.client ... --test --config v4/configs/ensemble_fused.json
+"""
+from __future__ import annotations
+
+import argparse
+import os
+import re
+import shutil
+import signal
+import socket
+import subprocess
+import sys
+import threading
+import time
+from pathlib import Path
+from typing import Optional
+
+import requests
+
+from .protocol import (
+    JobResult,
+    JobSpec,
+    PollResponse,
+    RegisterRequest,
+    RegisterResponse,
+    StatusPush,
+)
+
+# ──────────────────────────────────────────────────────────────
+# Server HTTP wrapper
+# ──────────────────────────────────────────────────────────────
+
+class _Server:
+    def __init__(self, base_url: str, token: str):
+        self.base_url = base_url.rstrip("/")
+        self._h = {"x-token": token}
+        self.client_id: str = ""
+
+    def _post(self, path: str, **kw) -> dict:
+        r = requests.post(f"{self.base_url}{path}", headers=self._h, timeout=15, **kw)
+        r.raise_for_status()
+        return r.json()
+
+    def register(self, hostname: str, gpu_info: str) -> str:
+        data = self._post("/register",
+                          json={"hostname": hostname, "gpu_info": gpu_info})
+        self.client_id = data["client_id"]
+        self.hostname = hostname
+        self.gpu_info = gpu_info
+        return self.client_id
+
+    def _reregister(self):
+        try:
+            self._post("/register",
+                       json={"hostname": self.hostname, "gpu_info": self.gpu_info},
+                       params={"reuse_id": self.client_id})
+        except Exception:
+            pass
+
+    def poll(self) -> Optional[JobSpec]:
+        data = self._post("/poll", params={"client_id": self.client_id})
+        if data.get("please_reregister"):
+            self._reregister()
+        return JobSpec(**data["job"]) if data.get("job") else None
+
+    def push_status(self, status: StatusPush):
+        try:
+            r = requests.post(
+                f"{self.base_url}/status/{self.client_id}",
+                json=status.model_dump(),
+                headers=self._h,
+                timeout=5,
+            )
+            if r.ok and r.json().get("please_reregister"):
+                self._reregister()
+        except Exception:
+            pass  # don't crash job on status push failure
+
+    def complete(self, job_id: str, success: bool, error_msg: Optional[str] = None):
+        self._post("/complete",
+                   json={"job_id": job_id, "success": success, "error_msg": error_msg})
+
+
+# ──────────────────────────────────────────────────────────────
+# GPU info
+# ──────────────────────────────────────────────────────────────
+
+def _gpu_info() -> str:
+    # NVIDIA
+    try:
+        out = subprocess.check_output(
+            ["nvidia-smi", "--query-gpu=name,memory.total", "--format=csv,noheader"],
+            text=True, stderr=subprocess.DEVNULL,
+        ).strip()
+        if out:
+            return " | ".join(out.splitlines())
+    except Exception:
+        pass
+    # AMD
+    try:
+        out = subprocess.check_output(
+            ["rocm-smi", "--showproductname", "--csv"],
+            text=True, stderr=subprocess.DEVNULL,
+        ).strip().splitlines()
+        names = [l for l in out if l and not l.startswith("device")]
+        if names:
+            return "AMD: " + " | ".join(names)
+    except Exception:
+        pass
+    # AMD fallback
+    try:
+        out = subprocess.check_output(
+            ["rocminfo"],
+            text=True, stderr=subprocess.DEVNULL,
+        )
+        names = [l.split(":", 1)[1].strip() for l in out.splitlines()
+                 if "Marketing Name:" in l]
+        if names:
+            return "AMD: " + " | ".join(names)
+    except Exception:
+        pass
+    return "no-gpu"
+
+
+# ──────────────────────────────────────────────────────────────
+# rsync helpers
+# ──────────────────────────────────────────────────────────────
+
+def _rsync(src: str, dst: str, delete: bool = False):
+    cmd = ["rsync", "-az", "--info=progress2"]
+    if delete:
+        cmd.append("--delete")
+    cmd += [src, dst]
+    subprocess.run(cmd, check=True)
+
+
+def _sync_code(server_ssh: str, server_path: str, local_path: str):
+    """Pull v4/ source from server → local (overwrites local changes)."""
+    src = f"{server_ssh}:{server_path}/v4/"
+    dst = f"{local_path}/v4/"
+    Path(dst).mkdir(parents=True, exist_ok=True)
+    _rsync(src, dst, delete=True)
+
+
+def _upload_results(server_ssh: str, server_path: str, local_path: str,
+                    run_name: str, output_dir: str):
+    src = f"{local_path}/{output_dir}/{run_name}/"
+    dst = f"{server_ssh}:{server_path}/{output_dir}/{run_name}/"
+    remote_parent = f"{server_path}/{output_dir}/{Path(run_name).parent}"
+    subprocess.run(["ssh", server_ssh, f"mkdir -p '{remote_parent}'"], check=True)
+    _rsync(src, dst)
+
+
+def _clean_local(local_path: str, run_name: str, output_dir: str):
+    target = Path(local_path) / output_dir / run_name
+    if target.exists():
+        shutil.rmtree(target)
+        print(f"[client] cleaned {target}", flush=True)
+
+
+# ──────────────────────────────────────────────────────────────
+# Stdout parsers (match v4_hypertower.py / fusion.py print format)
+# ──────────────────────────────────────────────────────────────
+
+# "  fold2 [nt] ep004/036 [fused_warmup  ]  loss=0.4321  acc=0.876  val_auc=0.7654"
+_EP_RE = re.compile(
+    r"fold(\d+)\s+\[([^\]]+)\]\s+ep(\d+)/(\d+).*?val_auc=([0-9.nan]+)",
+    re.IGNORECASE,
+)
+# "── fold 2/5  train_groups=..."
+_FOLD_RE = re.compile(r"fold\s+(\d+)/\d+", re.IGNORECASE)
+
+
+def _parse_line(line: str) -> dict:
+    """Return any structured fields found in a stdout line."""
+    out = {}
+    m = _FOLD_RE.search(line)
+    if m:
+        out["fold"] = int(m.group(1))
+    m = _EP_RE.search(line)
+    if m:
+        out["fold"]         = int(m.group(1))
+        out["stage"]        = m.group(2)
+        out["epoch"]        = int(m.group(3))
+        out["total_epochs"] = int(m.group(4))
+        try:
+            out["last_val_auc"] = float(m.group(5))
+        except ValueError:
+            pass
+    return out
+
+
+# ──────────────────────────────────────────────────────────────
+# Core job runner
+# ──────────────────────────────────────────────────────────────
+
+def _run_job(job: JobSpec, server: _Server,
+             server_ssh: str, server_path: str, local_path: str,
+             no_sync: bool = False,
+             extra_args: list[str] | None = None) -> bool:
+    extra_args = extra_args or []
+    # 1. Sync code
+    if no_sync:
+        print(f"[client] skipping sync (--no-sync)", flush=True)
+    else:
+        print(f"[client] syncing v4/ from server...", flush=True)
+        server.push_status(StatusPush(state="syncing", job_id=job.job_id, run_name=job.run_name))
+        _sync_code(server_ssh, server_path, local_path)
+
+    # 2. Launch training subprocess
+    cmd = [sys.executable, "-m", job.module] + job.args + extra_args
+    print(f"[client] running: {' '.join(cmd)}", flush=True)
+    server.push_status(StatusPush(state="running", job_id=job.job_id, run_name=job.run_name))
+
+    log_dir = Path(local_path) / "v4" / "distributed" / "logs"
+    log_dir.mkdir(parents=True, exist_ok=True)
+    log_file = log_dir / f"job_{job.job_id}.log"
+    ctx: dict = {}
+
+    def _tail(path: Path):
+        with open(path, "r") as f:
+            while True:
+                raw = f.readline()
+                if raw:
+                    print(raw, end="", flush=True)
+                    info = _parse_line(raw)
+                    ctx.update(info)
+                    if "epoch" in info:
+                        server.push_status(StatusPush(
+                            state="running",
+                            job_id=job.job_id,
+                            run_name=job.run_name,
+                            fold=ctx.get("fold"),
+                            stage=ctx.get("stage"),
+                            epoch=ctx.get("epoch"),
+                            total_epochs=ctx.get("total_epochs"),
+                            last_val_auc=ctx.get("last_val_auc"),
+                        ))
+                elif proc.poll() is not None:
+                    for raw in f:
+                        print(raw, end="", flush=True)
+                    break
+                else:
+                    time.sleep(0.05)
+
+    with open(log_file, "w") as logf:
+        proc = subprocess.Popen(
+            cmd,
+            stdout=logf,
+            stderr=logf,
+            cwd=local_path,
+            start_new_session=True,
+        )
+
+    tailer = threading.Thread(target=_tail, args=(log_file,), daemon=True)
+    tailer.start()
+    proc.wait()
+    tailer.join(timeout=5)
+    log_file.unlink(missing_ok=True)
+
+    success = proc.returncode == 0
+
+    if not success:
+        try:
+            os.killpg(os.getpgid(proc.pid), signal.SIGKILL)
+        except ProcessLookupError:
+            pass
+
+    if not success:
+        print(f"[client] job FAILED (rc={proc.returncode})", flush=True)
+        return False
+
+    # 3. Upload results
+    if no_sync:
+        print(f"[client] skipping upload (--no-sync, results already local)", flush=True)
+    else:
+        print(f"[client] uploading results...", flush=True)
+        server.push_status(StatusPush(state="uploading", job_id=job.job_id, run_name=job.run_name))
+        _upload_results(server_ssh, server_path, local_path, job.run_name, job.output_dir)
+        _clean_local(local_path, job.run_name, job.output_dir)
+
+    print(f"[client] job {job.job_id} complete.", flush=True)
+    return True
+
+
+# ──────────────────────────────────────────────────────────────
+# Compatibility test
+# ──────────────────────────────────────────────────────────────
+
+def _compat_test(server: _Server, server_ssh: str, server_path: str,
+                 local_path: str, config: str, extra_args: list[str],
+                 no_sync: bool = False):
+    """Run 1 fold to verify the env works end-to-end."""
+    print("[client] === compatibility test ===", flush=True)
+    job = JobSpec(
+        job_id="compat-test",
+        run_name="_compat_test",
+        module="v4.classes.v4_hypertower",
+        args=["--config", config, "--device", "cpu"] + extra_args,
+        output_dir="v4/results",
+    )
+    ok = _run_job(job, server, server_ssh, server_path, local_path,
+                  no_sync=no_sync)
+    _clean_local(local_path, "_compat_test", "v4/results")
+    if ok:
+        print("[client] compatibility test PASSED ✓", flush=True)
+    else:
+        print("[client] compatibility test FAILED ✗", flush=True)
+    return ok
+
+
+# ──────────────────────────────────────────────────────────────
+# Main daemon
+# ──────────────────────────────────────────────────────────────
+
+def main():
+    ap = argparse.ArgumentParser(
+        description=__doc__,
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+    ap.add_argument("--server", required=True,
+                    help="Server URL, e.g. http://apollo:8765")
+    ap.add_argument("--token", default=os.environ.get("HT_TOKEN", ""),
+                    help="Shared secret (or set HT_TOKEN env var)")
+    ap.add_argument("--server-ssh", required=True,
+                    help="SSH target for rsync, e.g. rpotter@apollo")
+    ap.add_argument("--server-path", required=True,
+                    help="Absolute path to hypertower root on server")
+    ap.add_argument("--local-path",
+                    default=str(Path.home() / "hypertower"),
+                    help="Absolute path to hypertower root on this machine")
+    ap.add_argument("--poll-interval", type=int, default=15,
+                    help="Seconds to wait between polls when idle")
+    ap.add_argument("--extra-args", nargs=argparse.REMAINDER, default=[],
+                    help="Extra args appended to every job on this client. "
+                         "Use -- to separate: --extra-args -- --device cpu")
+    ap.add_argument("--no-sync", action="store_true",
+                    help="Skip rsync of v4/ before each job (use when client IS the server)")
+    ap.add_argument("--test", action="store_true",
+                    help="Run 1-fold compatibility test and exit")
+    ap.add_argument("--config", default="v4/configs/ensemble_fused.json",
+                    help="Config path for --test mode")
+    args = ap.parse_args()
+
+    if not args.token:
+        ap.error("--token is required (or set HT_TOKEN)")
+
+    hostname = socket.gethostname()
+    gpu_info = _gpu_info()
+    server = _Server(args.server, args.token)
+
+    client_id = server.register(hostname, gpu_info)
+    print(f"[client] registered as {client_id} ({hostname} | {gpu_info})", flush=True)
+
+    if args.test:
+        extra = [a for a in args.extra_args if a != "--"]
+        sys.exit(0 if _compat_test(
+            server, args.server_ssh, args.server_path,
+            args.local_path, args.config, extra,
+            no_sync=args.no_sync,
+        ) else 1)
+
+    print(f"[client] polling every {args.poll_interval}s...", flush=True)
+    while True:
+        try:
+            job = server.poll()
+            if job is None:
+                server.push_status(StatusPush(state="idle"))
+                time.sleep(args.poll_interval)
+                continue
+
+            success = _run_job(
+                job, server,
+                args.server_ssh, args.server_path, args.local_path,
+                no_sync=args.no_sync,
+                extra_args=[a for a in args.extra_args if a != "--"],
+            )
+            server.complete(job.job_id, success,
+                            error_msg=None if success else "non-zero exit code")
+            server.push_status(StatusPush(state="idle"))
+
+        except KeyboardInterrupt:
+            print("\n[client] shutting down", flush=True)
+            break
+        except Exception as exc:
+            print(f"[client] error: {exc}", flush=True)
+            time.sleep(args.poll_interval)
+
+
+if __name__ == "__main__":
+    main()

v4/distributed/hypertower-server.service

@@ -0,0 +1,17 @@
+[Unit]
+Description=HyperTower distributed job server
+After=network-online.target
+Wants=network-online.target
+
+[Service]
+Type=simple
+WorkingDirectory=/home/rpotter/hypertower
+Environment=HT_TOKEN=hypertower
+ExecStart=/home/rpotter/miniconda3/envs/fundus_imaging/bin/python -m v4.distributed.server --host 0.0.0.0 --port 8765
+Restart=on-failure
+RestartSec=5
+StandardOutput=journal
+StandardError=journal
+
+[Install]
+WantedBy=default.target

v4/distributed/protocol.py

@@ -0,0 +1,61 @@
+"""Shared data models for server/client communication."""
+from __future__ import annotations
+
+from typing import Optional
+from pydantic import BaseModel
+
+
+class RegisterRequest(BaseModel):
+    hostname: str
+    gpu_info: str
+
+
+class RegisterResponse(BaseModel):
+    client_id: str
+
+
+class StatusPush(BaseModel):
+    state: str                          # idle | syncing | running | uploading | error
+    job_id: Optional[str] = None
+    run_name: Optional[str] = None
+    fold: Optional[int] = None
+    stage: Optional[str] = None
+    epoch: Optional[int] = None
+    total_epochs: Optional[int] = None
+    last_val_auc: Optional[float] = None
+    error: Optional[str] = None
+
+
+class ClientInfo(BaseModel):
+    client_id: str
+    hostname: str
+    gpu_info: str
+    status: StatusPush
+    last_seen: str
+
+
+class JobSpec(BaseModel):
+    job_id: str
+    run_name: str
+    module: str                         # e.g. "v4.classes.v4_hypertower"
+    args: list[str]
+    output_dir: str = "v4/results"
+
+
+class PollResponse(BaseModel):
+    job: Optional[JobSpec] = None
+    please_reregister: bool = False
+
+
+class JobResult(BaseModel):
+    job_id: str
+    success: bool
+    error_msg: Optional[str] = None
+
+
+class JobSubmit(BaseModel):
+    run_name: str
+    module: str = "v4.classes.v4_hypertower"
+    args: list[str]
+    output_dir: str = "v4/results"
+    priority: int = 0

v4/distributed/server.py

@@ -0,0 +1,476 @@
+"""
+HyperTower distributed job server.
+
+Manages a SQLite job queue and a registry of connected clients.
+Clients poll for work, push status updates, and report completion.
+
+Usage:
+    python -m v4.distributed.server --port 8765 --token <secret>
+
+Environment:
+    HT_TOKEN  — fallback if --token is not passed
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import os
+import sqlite3
+import threading
+import time
+import uuid
+from contextlib import contextmanager
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional
+
+from fastapi import Depends, FastAPI, Header, HTTPException
+import uvicorn
+
+from .protocol import (
+    ClientInfo,
+    JobResult,
+    JobSpec,
+    JobSubmit,
+    PollResponse,
+    RegisterRequest,
+    RegisterResponse,
+    StatusPush,
+)
+
+# ──────────────────────────────────────────────────────────────
+# Global state
+# ──────────────────────────────────────────────────────────────
+
+_TOKEN: str = ""
+_DB_PATH: Path = Path("v4/distributed/jobs.db")
+_REPO_ROOT: Path = Path.cwd()
+_CLIENT_TTL: int = 120  # seconds before a client is considered gone
+_MAX_ATTEMPTS: int = 3  # max times a job is retried before being left as failed
+
+_clients: dict[str, ClientInfo] = {}
+_clients_lock = threading.Lock()
+
+
+def _reap_stale_clients():
+    """Background thread: remove silent clients and re-queue their running jobs."""
+    while True:
+        time.sleep(30)
+        cutoff = datetime.now(timezone.utc).timestamp() - _CLIENT_TTL
+
+        with _clients_lock:
+            stale = [
+                cid
+                for cid, c in _clients.items()
+                if datetime.fromisoformat(c.last_seen).timestamp() < cutoff
+            ]
+            for cid in stale:
+                print(
+                    f"[server] reaped stale client {cid} ({_clients[cid].hostname})",
+                    flush=True,
+                )
+                del _clients[cid]
+            known_ids = set(_clients.keys())
+
+        with _db() as conn:
+            rows = conn.execute(
+                "SELECT job_id, assigned_to FROM jobs WHERE state='running'"
+            ).fetchall()
+            for row in rows:
+                if row["assigned_to"] not in known_ids:
+                    conn.execute(
+                        "UPDATE jobs SET state='pending', assigned_to=NULL, started_at=NULL "
+                        "WHERE job_id=?",
+                        (row["job_id"],),
+                    )
+                    print(
+                        f"[server] re-queued job {row['job_id']} "
+                        f"(client {row['assigned_to']} unknown)",
+                        flush=True,
+                    )
+
+
+# ──────────────────────────────────────────────────────────────
+# Database helpers
+# ──────────────────────────────────────────────────────────────
+
+
+@contextmanager
+def _db():
+    conn = sqlite3.connect(str(_DB_PATH))
+    conn.row_factory = sqlite3.Row
+    try:
+        yield conn
+        conn.commit()
+    finally:
+        conn.close()
+
+
+def _init_db():
+    _DB_PATH.parent.mkdir(parents=True, exist_ok=True)
+    with _db() as conn:
+        conn.execute(
+            """
+            CREATE TABLE IF NOT EXISTS jobs (
+                job_id       TEXT PRIMARY KEY,
+                run_name     TEXT NOT NULL,
+                module       TEXT NOT NULL,
+                args         TEXT NOT NULL,   -- JSON list
+                output_dir   TEXT NOT NULL DEFAULT 'v4/results',
+                state        TEXT NOT NULL DEFAULT 'pending',
+                priority     INTEGER NOT NULL DEFAULT 0,
+                assigned_to  TEXT,
+                created_at   TEXT NOT NULL,
+                started_at   TEXT,
+                completed_at TEXT,
+                error_msg    TEXT,
+                attempts     INTEGER NOT NULL DEFAULT 0
+            )
+        """
+        )
+        conn.execute("CREATE INDEX IF NOT EXISTS idx_jobs_state ON jobs(state)")
+        conn.execute("CREATE INDEX IF NOT EXISTS idx_jobs_priority ON jobs(priority)")
+        conn.execute("CREATE INDEX IF NOT EXISTS idx_jobs_run_name ON jobs(run_name)")
+        conn.execute("CREATE UNIQUE INDEX IF NOT EXISTS idx_jobs_args ON jobs(args)")
+        try:
+            conn.execute(
+                "ALTER TABLE jobs ADD COLUMN attempts INTEGER NOT NULL DEFAULT 0"
+            )
+        except Exception:
+            pass  # column already exists
+
+
+def _now() -> str:
+    return datetime.now(timezone.utc).isoformat()
+
+
+def _ensure_client(client_id: str, hostname: str = "", gpu_info: str = "") -> bool:
+    """Re-register a client that survived a server restart.
+    Returns True if the client was unknown (placeholder created)."""
+    if client_id not in _clients:
+        _clients[client_id] = ClientInfo(
+            client_id=client_id,
+            hostname=hostname or client_id,
+            gpu_info=gpu_info or "unknown",
+            status=StatusPush(state="idle"),
+            last_seen=_now(),
+        )
+        print(f"[server] re-registered {client_id} (survived restart)", flush=True)
+        return True
+    return False
+
+
+# ──────────────────────────────────────────────────────────────
+# FastAPI app
+# ──────────────────────────────────────────────────────────────
+
+app = FastAPI(title="HyperTower Job Server")
+
+
+def _check_token(x_token: str = Header(...)):
+    if x_token != _TOKEN:
+        raise HTTPException(status_code=403, detail="Invalid token")
+
+
+# ── Registration ──────────────────────────────────────────────
+
+
+@app.post(
+    "/register", response_model=RegisterResponse, dependencies=[Depends(_check_token)]
+)
+def register(req: RegisterRequest, reuse_id: Optional[str] = None):
+    with _clients_lock:
+        client_id = (
+            reuse_id if (reuse_id and reuse_id in _clients) else str(uuid.uuid4())[:8]
+        )
+        existing_status = (
+            _clients[client_id].status
+            if client_id in _clients
+            else StatusPush(state="idle")
+        )
+        _clients[client_id] = ClientInfo(
+            client_id=client_id,
+            hostname=req.hostname,
+            gpu_info=req.gpu_info,
+            status=existing_status,
+            last_seen=_now(),
+        )
+    action = "re-registered" if reuse_id else "registered"
+    print(
+        f"[server] {action} {client_id} ({req.hostname} | {req.gpu_info})", flush=True
+    )
+    return RegisterResponse(client_id=client_id)
+
+
+# ── Job polling ───────────────────────────────────────────────
+
+
+@app.post("/poll", response_model=PollResponse, dependencies=[Depends(_check_token)])
+def poll(client_id: str):
+    with _clients_lock:
+        needs_reregister = _ensure_client(client_id)
+        _clients[client_id].last_seen = _now()
+
+    with _db() as conn:
+        row = conn.execute(
+            "SELECT * FROM jobs WHERE state='pending' "
+            "ORDER BY priority DESC, created_at ASC LIMIT 1"
+        ).fetchone()
+
+        if row is None:
+            return PollResponse(job=None, please_reregister=needs_reregister)
+
+        job_id = row["job_id"]
+        cur = conn.execute(
+            "UPDATE jobs SET state='pending', assigned_to=NULL, started_at=NULL "
+            "WHERE assigned_to=? AND state='running' AND job_id!=?",
+            (client_id, job_id),
+        )
+        if cur.rowcount:
+            print(
+                f"[server] reset {cur.rowcount} orphaned running job(s) for {client_id}",
+                flush=True,
+            )
+        conn.execute(
+            "UPDATE jobs SET state='running', assigned_to=?, started_at=? WHERE job_id=?",
+            (client_id, _now(), job_id),
+        )
+
+    job = JobSpec(
+        job_id=job_id,
+        run_name=row["run_name"],
+        module=row["module"],
+        args=json.loads(row["args"]),
+        output_dir=row["output_dir"],
+    )
+
+    with _clients_lock:
+        _clients[client_id].status = StatusPush(
+            state="syncing", job_id=job_id, run_name=row["run_name"]
+        )
+
+    print(f"[server] dispatched {job_id} ({row['run_name']}) → {client_id}", flush=True)
+    return PollResponse(job=job, please_reregister=needs_reregister)
+
+
+# ── Status ────────────────────────────────────────────────────
+
+
+@app.post("/status/{client_id}", dependencies=[Depends(_check_token)])
+def push_status(client_id: str, status: StatusPush):
+    with _clients_lock:
+        needs_reregister = _ensure_client(client_id)
+        _clients[client_id].status = status
+        _clients[client_id].last_seen = _now()
+    return {"ok": True, "please_reregister": needs_reregister}
+
+
+@app.get("/clients", dependencies=[Depends(_check_token)])
+def list_clients():
+    with _clients_lock:
+        return list(_clients.values())
+
+
+@app.get("/clients/{client_id}", dependencies=[Depends(_check_token)])
+def get_client(client_id: str):
+    with _clients_lock:
+        if client_id not in _clients:
+            raise HTTPException(status_code=404, detail="Unknown client")
+        return _clients[client_id]
+
+
+# ── Job completion ────────────────────────────────────────────
+
+
+@app.post("/complete", dependencies=[Depends(_check_token)])
+def complete(result: JobResult):
+    with _db() as conn:
+        if result.success:
+            conn.execute(
+                "UPDATE jobs SET state='done', completed_at=?, error_msg=NULL WHERE job_id=?",
+                (_now(), result.job_id),
+            )
+            print(f"[server] job {result.job_id} → done", flush=True)
+
+            run_row = conn.execute(
+                "SELECT run_name FROM jobs WHERE job_id=?", (result.job_id,)
+            ).fetchone()
+            if run_row:
+                run_name = run_row["run_name"]
+                remaining = conn.execute(
+                    "SELECT COUNT(*) FROM jobs WHERE run_name=? AND state != 'done'",
+                    (run_name,),
+                ).fetchone()[0]
+                if remaining == 0:
+                    total = conn.execute(
+                        "SELECT COUNT(*) FROM jobs WHERE run_name=?", (run_name,)
+                    ).fetchone()[0]
+                    conn.execute("DELETE FROM jobs WHERE run_name=?", (run_name,))
+                    print(
+                        f"[server] run '{run_name}' complete ({total} jobs) — cleared",
+                        flush=True,
+                    )
+        else:
+            row = conn.execute(
+                "SELECT attempts FROM jobs WHERE job_id=?", (result.job_id,)
+            ).fetchone()
+            attempts = (row["attempts"] if row else 0) + 1
+            if attempts < _MAX_ATTEMPTS:
+                conn.execute(
+                    "UPDATE jobs SET state='pending', assigned_to=NULL, started_at=NULL, "
+                    "attempts=?, error_msg=? WHERE job_id=?",
+                    (attempts, result.error_msg, result.job_id),
+                )
+                print(
+                    f"[server] job {result.job_id} failed (attempt {attempts}/{_MAX_ATTEMPTS}), "
+                    f"re-queuing",
+                    flush=True,
+                )
+            else:
+                conn.execute(
+                    "UPDATE jobs SET state='failed', completed_at=?, attempts=?, error_msg=? "
+                    "WHERE job_id=?",
+                    (_now(), attempts, result.error_msg, result.job_id),
+                )
+                print(
+                    f"[server] job {result.job_id} failed permanently after "
+                    f"{attempts} attempts",
+                    flush=True,
+                )
+    return {"ok": True}
+
+
+# ── Job queue management ──────────────────────────────────────
+
+
+@app.post("/jobs", dependencies=[Depends(_check_token)])
+def submit_job(job: JobSubmit):
+    result_dir = _REPO_ROOT / job.output_dir / job.run_name
+    if result_dir.exists() and any(result_dir.rglob("summary.json")):
+        print(f"[server] skipped {job.run_name} (results exist on disk)", flush=True)
+        return {"job_id": "", "duplicate": False, "skipped": True}
+
+    args_json = json.dumps(job.args)
+    job_id    = str(uuid.uuid4())[:12]
+    with _db() as conn:
+        cur = conn.execute(
+            "INSERT OR IGNORE INTO jobs "
+            "(job_id, run_name, module, args, output_dir, priority, created_at) "
+            "VALUES (?,?,?,?,?,?,?)",
+            (job_id, job.run_name, job.module, args_json,
+             job.output_dir, job.priority, _now()),
+        )
+        if cur.rowcount == 0:
+            existing = conn.execute(
+                "SELECT job_id FROM jobs WHERE args=?", (args_json,)
+            ).fetchone()
+            job_id = existing["job_id"]
+            print(f"[server] duplicate ignored ({job.run_name}) → {job_id}", flush=True)
+            return {"job_id": job_id, "duplicate": True}
+    print(f"[server] queued {job_id} ({job.run_name})", flush=True)
+    return {"job_id": job_id, "duplicate": False}
+
+
+@app.get("/jobs", dependencies=[Depends(_check_token)])
+def list_jobs(state: Optional[str] = None):
+    with _db() as conn:
+        if state:
+            rows = conn.execute(
+                "SELECT * FROM jobs WHERE state=? ORDER BY created_at DESC", (state,)
+            ).fetchall()
+        else:
+            rows = conn.execute(
+                "SELECT * FROM jobs ORDER BY created_at DESC"
+            ).fetchall()
+    return [dict(r) for r in rows]
+
+
+@app.post("/jobs/clear", dependencies=[Depends(_check_token)])
+def clear_jobs(body: dict):
+    with _db() as conn:
+        if body.get("all"):
+            cur = conn.execute("DELETE FROM jobs")
+        elif body.get("run_name"):
+            cur = conn.execute("DELETE FROM jobs WHERE run_name=?", (body["run_name"],))
+        else:
+            states = body.get("states", ["done", "failed", "cancelled"])
+            placeholders = ",".join("?" * len(states))
+            cur = conn.execute(
+                f"DELETE FROM jobs WHERE state IN ({placeholders})", states
+            )
+    print(f"[server] cleared {cur.rowcount} jobs", flush=True)
+    return {"cleared": cur.rowcount}
+
+
+@app.delete("/jobs/{job_id}", dependencies=[Depends(_check_token)])
+def cancel_job(job_id: str):
+    with _db() as conn:
+        conn.execute(
+            "UPDATE jobs SET state='cancelled' WHERE job_id=? AND state='pending'",
+            (job_id,),
+        )
+    return {"ok": True}
+
+
+# ──────────────────────────────────────────────────────────────
+# Entry point
+# ──────────────────────────────────────────────────────────────
+
+
+def main():
+    ap = argparse.ArgumentParser(
+        description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter
+    )
+    ap.add_argument("--port", type=int, default=8765)
+    ap.add_argument("--host", default="0.0.0.0")
+    ap.add_argument(
+        "--token",
+        default=os.environ.get("HT_TOKEN", ""),
+        help="Shared secret (or set HT_TOKEN env var)",
+    )
+    ap.add_argument(
+        "--db", default="v4/distributed/jobs.db", help="Path to SQLite job database"
+    )
+    ap.add_argument(
+        "--client-ttl",
+        type=int,
+        default=120,
+        help="Seconds of silence before a client is reaped (default: 120)",
+    )
+    ap.add_argument(
+        "--max-attempts",
+        type=int,
+        default=3,
+        help="Max times a failed job is retried before being left as failed (default: 3)",
+    )
+    ap.add_argument(
+        "--root",
+        default="",
+        help="Repo root for results-existence checks (default: cwd)",
+    )
+    args = ap.parse_args()
+
+    if not args.token:
+        ap.error("--token is required (or set HT_TOKEN)")
+
+    global _TOKEN, _DB_PATH, _REPO_ROOT, _CLIENT_TTL, _MAX_ATTEMPTS
+    _TOKEN = args.token
+    _DB_PATH = Path(args.db)
+    _REPO_ROOT = Path(args.root).resolve() if args.root else Path.cwd()
+    _CLIENT_TTL = args.client_ttl
+    _MAX_ATTEMPTS = args.max_attempts
+    _init_db()
+
+    reaper = threading.Thread(target=_reap_stale_clients, daemon=True)
+    reaper.start()
+
+    print(
+        f"[server] listening on {args.host}:{args.port}  client_ttl={_CLIENT_TTL}s",
+        flush=True,
+    )
+    uvicorn.run(app, host=args.host, port=args.port, log_level="warning")
+
+
+if __name__ == "__main__":
+    main()

v4/distributed/server_cheat_sheet.md

@@ -0,0 +1,147 @@
+# Distributed Server Cheat Sheet
+
+All commands assume server is running on hades at port 8765.
+
+## Start Server
+```bash
+python -m v4.distributed.server --token hypertower
+```
+Run inside tmux so it survives disconnects:
+```bash
+tmux new -s htserver
+python -m v4.distributed.server --token hypertower
+# Ctrl-B D to detach
+tmux attach -t htserver   # reattach later
+```
+
+## Local Workflow (hades as server + client)
+
+```bash
+# Terminal 1 — server
+python -m v4.distributed.server --token hypertower
+
+# Terminal 2 — client (skip rsync, results already local)
+python -m v4.distributed.client \
+    --server http://hades:8765 --token hypertower \
+    --server-ssh ignored --server-path ignored \
+    --local-path /home/rpotter/hypertower \
+    --no-sync
+
+# Terminal 3 — dispatch (exits after queuing; client picks up jobs)
+python -m v4.distributed.batch_dispatch \
+    --server http://hades:8765 --token hypertower \
+    --config v4/configs/ensemble_fused.json \
+    --batch v4/scripts/experiments/my_batch.json
+```
+
+---
+
+## Start Clients
+
+**Hades (server-local, no sync):**
+```bash
+python -m v4.distributed.client \
+    --server http://hades:8765 --token hypertower \
+    --server-ssh ignored --server-path ignored \
+    --local-path /home/rpotter/hypertower \
+    --no-sync
+```
+
+**Apollo (remote client):**
+```bash
+python -m v4.distributed.client \
+    --server http://hades:8765 --token hypertower \
+    --server-ssh rpotter@hades \
+    --server-path /home/rpotter/hypertower \
+    --local-path /home/odin/hypertower
+```
+
+---
+
+## Monitoring
+
+**Live client monitor (refreshes every 5s):**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clients --watch
+```
+
+**Faster refresh:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clients --watch --interval 2
+```
+
+**Inspect a single client:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clients <client_id>
+```
+
+**View job queue:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower jobs
+```
+
+**Filter by state:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower jobs --state pending
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower jobs --state running
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower jobs --state failed
+```
+
+---
+
+## Submitting Jobs
+
+**Batch dispatch (dry run first):**
+```bash
+python -m v4.distributed.batch_dispatch \
+    --server http://hades:8765 --token hypertower \
+    --config v4/configs/ensemble_fused.json \
+    --batch v4/scripts/experiments/fusion_dim_sweep.json \
+    --dry-run
+```
+
+**Live submit:**
+```bash
+python -m v4.distributed.batch_dispatch \
+    --server http://hades:8765 --token hypertower \
+    --config v4/configs/ensemble_fused.json \
+    --batch v4/scripts/experiments/fusion_dim_sweep.json
+```
+
+**Fewer reps (e.g. quick test):**
+```bash
+python -m v4.distributed.batch_dispatch \
+    --server http://hades:8765 --token hypertower \
+    --config v4/configs/ensemble_fused.json \
+    --batch v4/scripts/experiments/fusion_dim_sweep.json \
+    --reps 3
+```
+
+---
+
+## Queue Management
+
+**Clear failed jobs:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clear --states failed
+```
+
+**Clear running jobs (orphan cleanup):**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clear --states running
+```
+
+**Clear all jobs:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clear --all
+```
+
+**Clear by run name:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower clear --run-name experiments/fusion_dim_sweep/dim256
+```
+
+**Cancel a specific job:**
+```bash
+python -m v4.distributed.cli --server http://hades:8765 --token hypertower cancel <job_id>
+```

v4/results/v4/ensemble_fused/binary/ntower/gold_standard.json

@@ -1,146 +0,0 @@
-{
-  "run_name": "v4/ensemble_fused",
-  "config": {
-    "_notes": [
-      "V4 ensemble_fused: img + cd towers, HTFusion Stage 1, HyperBridge Stage 2.",
-      "Matches phase5/embedding_mlp_head setup for direct comparison.",
-      "data_source resolves against the PapilaBundle returned by build_data.",
-      "image_dir / clinical_dir are project-relative; orchestrator resolves against REPO_ROOT."
-    ],
-    "run_name": "v4/ensemble_fused",
-    "eval_mode": "binary",
-    "split_identity_level": 1,
-    "epochs": 30,
-    "fusion_epochs": 10,
-    "folds": 5,
-    "fold_seed": 100,
-    "seed": 1234,
-    "output_root": "v4/results",
-    "data": {
-      "module": "v4.classes.profiles.v4papila",
-      "args": {
-        "image_dir": "Papila/FundusImages",
-        "clinical_dir": "Papila/ClinicalData",
-        "label_col": "Diagnosis",
-        "iop_corr_method": "ratio",
-        "iop_drop_raw": true,
-        "exclude_cols": [
-          "Axial_Length"
-        ],
-        "in_memory_cache": true
-      }
-    },
-    "towers": [
-      {
-        "name": "img",
-        "module": "v4.classes.towers.image_tower",
-        "class": "ImageEncoder",
-        "data_source": "image",
-        "args": {
-          "backbone": "refugelike",
-          "freeze_ratio": 0.0,
-          "augment": true
-        },
-        "warmup_epochs": 0
-      },
-      {
-        "name": "cd",
-        "module": "v4.classes.towers.clinical_tower",
-        "class": "ClinicalEncoder",
-        "data_source": "matrix",
-        "args": {
-          "hidden_dim": 128
-        },
-        "warmup_epochs": 40,
-        "warmup_exclude_towers": [
-          "img"
-        ]
-      }
-    ],
-    "bridge": {
-      "mode": "embedding_mlp",
-      "fusion_dim": 256,
-      "hidden_dim": 256
-    },
-    "training": {
-      "lr": 0.0001,
-      "batch_size": 16,
-      "bcd_prob": 0.5,
-      "warmup_tower_epochs": 3,
-      "warmup_fused_epochs": 3,
-      "tune_binary_threshold": true
-    }
-  },
-  "mean_val_auc": 0.9036764705882353,
-  "std_val_auc": 0.02318218054475656,
-  "mean_test_auc": 0.9007352941176471,
-  "std_test_auc": 0.04178903599524356,
-  "elapsed_s": 1563.1,
-  "fold_results": [
-    {
-      "fold": 0,
-      "val_auc": 0.9044117647058824,
-      "val_acc": 0.8571428571428571,
-      "val_n": 42,
-      "val_kappa": 0.42727272727272725,
-      "val_mcc": 0.4622975667767223,
-      "val_f1": 0.7083333333333333,
-      "val_threshold": 0.1659889668226242,
-      "test_auc": 0.8566176470588236,
-      "test_acc": 0.9285714285714286,
-      "test_n": 42
-    },
-    {
-      "fold": 1,
-      "val_auc": 0.8970588235294118,
-      "val_acc": 0.8809523809523809,
-      "val_n": 42,
-      "val_kappa": 0.5945945945945946,
-      "val_mcc": 0.5965587590013045,
-      "val_f1": 0.7971014492753623,
-      "val_threshold": 0.017083797603845596,
-      "test_auc": 0.9044117647058824,
-      "test_acc": 0.8095238095238095,
-      "test_n": 42
-    },
-    {
-      "fold": 2,
-      "val_auc": 0.863970588235294,
-      "val_acc": 0.8333333333333334,
-      "val_n": 42,
-      "val_kappa": 0.43243243243243246,
-      "val_mcc": 0.4338609156373123,
-      "val_f1": 0.7159420289855072,
-      "val_threshold": 0.09843172132968903,
-      "test_auc": 0.9044117647058824,
-      "test_acc": 0.8571428571428571,
-      "test_n": 42
-    },
-    {
-      "fold": 3,
-      "val_auc": 0.9227941176470588,
-      "val_acc": 0.8809523809523809,
-      "val_n": 42,
-      "val_kappa": 0.631578947368421,
-      "val_mcc": 0.6333004963811236,
-      "val_f1": 0.8156277436347674,
-      "val_threshold": 0.01531070377677679,
-      "test_auc": 0.8639705882352942,
-      "test_acc": 0.8571428571428571,
-      "test_n": 42
-    },
-    {
-      "fold": 4,
-      "val_auc": 0.9301470588235294,
-      "val_acc": 0.9047619047619048,
-      "val_n": 42,
-      "val_kappa": 0.6181818181818182,
-      "val_mcc": 0.6688560540599386,
-      "val_f1": 0.8055555555555556,
-      "val_threshold": 0.19587548077106476,
-      "test_auc": 0.9742647058823529,
-      "test_acc": 0.9285714285714286,
-      "test_n": 42
-    }
-  ]
-}

v4/results/v4/ensemble_fused/binary/ntower/summary.json

@@ -1,264 +0,0 @@
-{
-  "run_name": "v4/ensemble_fused",
-  "eval_stage": "hb",
-  "config": {
-    "_notes": [
-      "V4 stage-pipeline: warm \u2192 fusion stages with parallel head stages.",
-      "Bridges are pure embedding producers; heads are separate swappable stages.",
-      "BCD-eligible heads are sampled during tower_warmup and main phases.",
-      "eval_stage names the fusion stage whose primary head is used for final metrics."
-    ],
-    "run_name": "v4/ensemble_fused",
-    "num_classes": 2,
-    "label_filter": [
-      0,
-      1
-    ],
-    "split_identity_level": 1,
-    "eval_stage": "hb",
-    "save_predictions": false,
-    "seed": 1234,
-    "folds": 5,
-    "fold_seed": 100,
-    "output_root": "v4/results",
-    "out_dir_tags": [
-      "binary",
-      "ntower"
-    ],
-    "data": {
-      "module": "v4.classes.profiles.v4papila",
-      "args": {
-        "image_dir": "Papila/FundusImages",
-        "clinical_dir": "Papila/ClinicalData",
-        "label_col": "Diagnosis",
-        "iop_corr_method": "ratio",
-        "iop_drop_raw": true,
-        "exclude_cols": [
-          "Axial_Length"
-        ],
-        "in_memory_cache": true
-      }
-    },
-    "towers": [
-      {
-        "name": "img",
-        "module": "v4.classes.towers.image_tower",
-        "class": "ImageEncoder",
-        "data_source": "image",
-        "args": {
-          "backbone": "refugelike",
-          "freeze_ratio": 0.0,
-          "augment": true
-        }
-      },
-      {
-        "name": "cd",
-        "module": "v4.classes.towers.clinical_tower",
-        "class": "ClinicalEncoder",
-        "data_source": "matrix",
-        "args": {
-          "hidden_dim": 128
-        }
-      }
-    ],
-    "stages": [
-      {
-        "name": "cd_warm",
-        "type": "warm",
-        "tower": "cd",
-        "level": "eye",
-        "epochs": 40
-      },
-      {
-        "name": "img_aux",
-        "type": "head",
-        "input": "img",
-        "train_with": "nt",
-        "bcd": true
-      },
-      {
-        "name": "cd_aux",
-        "type": "head",
-        "input": "cd",
-        "train_with": "nt",
-        "bcd": true
-      },
-      {
-        "name": "nt",
-        "type": "fusion",
-        "module": "v4.classes.bridges.fusion_bridge",
-        "class": "FusionBridge",
-        "inputs": [
-          "img",
-          "cd"
-        ],
-        "level": "eye",
-        "epochs": 36,
-        "train_towers": true,
-        "warmup": {
-          "tower_epochs": 3,
-          "fused_epochs": 3
-        },
-        "args": {
-          "fusion_dim": 256
-        }
-      },
-      {
-        "name": "nt_head",
-        "type": "head",
-        "input": "nt",
-        "train_with": "nt"
-      },
-      {
-        "name": "hb",
-        "type": "fusion",
-        "module": "v4.classes.bridges.hyperbridge",
-        "class": "HyperBridge",
-        "inputs": {
-          "a": "nt",
-          "b": "nt"
-        },
-        "level": "patient",
-        "epochs": 10,
-        "args": {
-          "hidden_dim": 256,
-          "mode": "embedding_mlp"
-        }
-      },
-      {
-        "name": "hb_head",
-        "type": "head",
-        "input": "hb",
-        "train_with": "hb"
-      }
-    ],
-    "training": {
-      "lr": 0.0001,
-      "batch_size": 16,
-      "bcd_prob": 0.5,
-      "tune_binary_threshold": true
-    }
-  },
-  "mean_val_auc": 0.8926470588235293,
-  "std_val_auc": 0.06105155516072669,
-  "mean_test_auc": 0.9022058823529413,
-  "std_test_auc": 0.035001853633564395,
-  "elapsed_s": 1525.9,
-  "fold_results": [
-    {
-      "fold": 0,
-      "nt_val_auc": 0.858647936786655,
-      "nt_val_acc": 0.8928571428571429,
-      "nt_val_n": 84,
-      "nt_val_kappa": 0.6272189349112426,
-      "nt_val_mcc": 0.6411186083279721,
-      "nt_val_f1": 0.8124534854874721,
-      "nt_val_threshold": 0.012987074442207813,
-      "nt_test_auc": 0.9157155399473222,
-      "nt_test_acc": 0.9047619047619048,
-      "nt_test_n": 84,
-      "hb_val_auc": 0.9080882352941178,
-      "hb_val_acc": 0.8333333333333334,
-      "hb_val_n": 42,
-      "hb_val_kappa": 0.2898550724637682,
-      "hb_val_mcc": 0.33633639699815626,
-      "hb_val_f1": 0.6338729763387297,
-      "hb_val_threshold": 0.011891158297657967,
-      "hb_test_auc": 0.9522058823529411,
-      "hb_test_acc": 0.9523809523809523,
-      "hb_test_n": 42
-    },
-    {
-      "fold": 1,
-      "nt_val_auc": 0.8282828282828283,
-      "nt_val_acc": 0.8928571428571429,
-      "nt_val_n": 84,
-      "nt_val_kappa": 0.6111111111111112,
-      "nt_val_mcc": 0.6633249580710799,
-      "nt_val_f1": 0.801418439716312,
-      "nt_val_threshold": 0.008082838729023933,
-      "nt_test_auc": 0.8726953467954346,
-      "nt_test_acc": 0.8690476190476191,
-      "nt_test_n": 84,
-      "hb_val_auc": 0.7904411764705882,
-      "hb_val_acc": 0.9047619047619048,
-      "hb_val_n": 42,
-      "hb_val_kappa": 0.6181818181818182,
-      "hb_val_mcc": 0.6688560540599386,
-      "hb_val_f1": 0.8055555555555556,
-      "hb_val_threshold": 0.013159404508769512,
-      "hb_test_auc": 0.8970588235294117,
-      "hb_test_acc": 0.8571428571428571,
-      "hb_test_n": 42
-    },
-    {
-      "fold": 2,
-      "nt_val_auc": 0.8906882591093117,
-      "nt_val_acc": 0.8452380952380952,
-      "nt_val_n": 84,
-      "nt_val_kappa": 0.5125,
-      "nt_val_mcc": 0.5217535056401378,
-      "nt_val_f1": 0.7548821548821549,
-      "nt_val_threshold": 0.052708517760038376,
-      "nt_test_auc": 0.856060606060606,
-      "nt_test_acc": 0.8333333333333334,
-      "nt_test_n": 84,
-      "hb_val_auc": 0.9117647058823529,
-      "hb_val_acc": 0.8333333333333334,
-      "hb_val_n": 42,
-      "hb_val_kappa": 0.36909871244635195,
-      "hb_val_mcc": 0.3833788364965519,
-      "hb_val_f1": 0.6814734561213435,
-      "hb_val_threshold": 0.0401872955262661,
-      "hb_test_auc": 0.8566176470588236,
-      "hb_test_acc": 0.8571428571428571,
-      "hb_test_n": 42
-    },
-    {
-      "fold": 3,
-      "nt_val_auc": 0.921875,
-      "nt_val_acc": 0.8928571428571429,
-      "nt_val_n": 84,
-      "nt_val_kappa": 0.6758147512864494,
-      "nt_val_mcc": 0.6797955088067001,
-      "nt_val_f1": 0.837593984962406,
-      "nt_val_threshold": 0.708580732345581,
-      "nt_test_auc": 0.8631578947368421,
-      "nt_test_acc": 0.8452380952380952,
-      "nt_test_n": 84,
-      "hb_val_auc": 0.9779411764705882,
-      "hb_val_acc": 0.9285714285714286,
-      "hb_val_n": 42,
-      "hb_val_kappa": 0.7567567567567568,
-      "hb_val_mcc": 0.7592566023652966,
-      "hb_val_f1": 0.8782608695652174,
-      "hb_val_threshold": 0.04653067886829376,
-      "hb_test_auc": 0.875,
-      "hb_test_acc": 0.8809523809523809,
-      "hb_test_n": 42
-    },
-    {
-      "fold": 4,
-      "nt_val_auc": 0.8279192273924496,
-      "nt_val_acc": 0.8571428571428571,
-      "nt_val_n": 84,
-      "nt_val_kappa": 0.46325878594249204,
-      "nt_val_mcc": 0.49610717544581684,
-      "nt_val_f1": 0.7269772481040087,
-      "nt_val_threshold": 0.06419441103935242,
-      "nt_test_auc": 0.8961397058823529,
-      "nt_test_acc": 0.8928571428571429,
-      "nt_test_n": 84,
-      "hb_val_auc": 0.875,
-      "hb_val_acc": 0.9047619047619048,
-      "hb_val_n": 42,
-      "hb_val_kappa": 0.6181818181818182,
-      "hb_val_mcc": 0.6688560540599386,
-      "hb_val_f1": 0.8055555555555556,
-      "hb_val_threshold": 0.23509082198143005,
-      "hb_test_auc": 0.9301470588235294,
-      "hb_test_acc": 0.8809523809523809,
-      "hb_test_n": 42
-    }
-  ]
-}

v4/scripts/experiments/geom_gt_sweep.json

@@ -0,0 +1,5 @@
+[
+  {
+    "run_name": "experiments/ensemble_fused/geom_gt"
+  }
+]

v4/scripts/experiments/geometry_vec_gt_sweep.json

@@ -0,0 +1,6 @@
+[
+  {
+    "_note": "Geometry vector injection, GT source, all 5 features — v4 equivalent of v3 phase6a.",
+    "run_name": "experiments/geometry_vec_gt/dim5"
+  }
+]

v4/scripts/experiments/no_geom_sweep.json

@@ -0,0 +1,5 @@
+[
+  {
+    "run_name": "experiments/ensemble_fused/no_geom"
+  }
+]

v4/scripts/experiments/tri_v1/baseline_ensemble.json

@@ -0,0 +1,6 @@
+[
+  {
+    "_note": "Replacement run of ensemble_fused (img + cd) — keeps original alongside.",
+    "run_name": "experiments/tri_v1/baseline_ensemble"
+  }
+]

v4/scripts/experiments/tri_v1/baseline_solo.json

@@ -0,0 +1,6 @@
+[
+  {
+    "_note": "Standalone geometry tower (UNet seg, MonoBridge) — 10-rep baseline.",
+    "run_name": "experiments/tri_v1/baseline_solo"
+  }
+]

v4/scripts/experiments/tri_v1/baseline_tri.json

@@ -0,0 +1,6 @@
+[
+  {
+    "_note": "Tritower with default params (bcd_prob=0.5, cw=false, nt_epochs=36) — 10-rep baseline.",
+    "run_name": "experiments/tri_v1/baseline_tri"
+  }
+]

v4/scripts/experiments/tri_v1/grid.json

@@ -0,0 +1,22 @@
+[
+  { "_note": "tritower grid pilot (3 reps each) — 3 bcd × 2 cw × 3 nt_epochs = 18 cells",   "run_name": "experiments/tri_v1/grid/bcd35_cw0_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd35_cw0_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd35_cw0_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 36 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd35_cw1_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd35_cw1_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd35_cw1_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.35, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 36 } } },
+
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw0_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw0_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw0_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 36 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw1_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw1_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd50_cw1_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.50, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 36 } } },
+
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw0_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw0_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw0_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": false } }, "stage_overrides": { "nt": { "epochs": 36 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw1_nt15", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 15 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw1_nt25", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 25 } } },
+  { "run_name": "experiments/tri_v1/grid/bcd75_cw1_nt36", "reps": 3, "overrides": { "training": { "bcd_prob": 0.75, "class_weighted": true  } }, "stage_overrides": { "nt": { "epochs": 36 } } }
+]