diff --git a/docs/superpowers/plans/2026-04-10-argo-pipeline-redesign.md b/docs/superpowers/plans/2026-04-10-argo-pipeline-redesign.md
new file mode 100644
index 000000000..57e398804
--- /dev/null
+++ b/docs/superpowers/plans/2026-04-10-argo-pipeline-redesign.md
@@ -0,0 +1,647 @@
+# Argo Training Pipeline Redesign Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Replace 7 overlapping Argo workflow templates with a single unified `train` template that caches binaries by commit SHA on PVC and skips fxcache regeneration when valid.
+
+**Architecture:** One workflow template (`train-template.yaml`) with a DAG: `ensure-binary` (cache-or-compile) and `gpu-warmup` run in parallel, then `ensure-fxcache` (cache-or-precompute), then `hyperopt` → `train-best` → `evaluate` → `upload-results`. One script (`argo-train.sh`) submits it. Five dead templates deleted.
+
+**Tech Stack:** Argo Workflows, Kubernetes, bash, PVC storage
+
+---
+
+## File Structure
+
+| Action | File | Responsibility |
+|--------|------|----------------|
+| Create | `infra/k8s/argo/train-template.yaml` | Unified training workflow template |
+| Rewrite | `scripts/argo-train.sh` | Simplified CLI that submits the unified template |
+| Delete | `infra/k8s/argo/compile-and-train-template.yaml` | Replaced by train-template |
+| Delete | `infra/k8s/argo/train-dqn-template.yaml` | Unused duplicate |
+| Delete | `infra/k8s/argo/train-baseline-rl-template.yaml` | Unused duplicate |
+| Delete | `infra/k8s/argo/train-supervised-template.yaml` | Unused duplicate |
+| Delete | `infra/k8s/argo/training-workflow-template.yaml` | Unused duplicate |
+| Delete | `infra/k8s/argo/precompute-features-template.yaml` | Absorbed into ensure-fxcache step |
+| Delete | `scripts/argo-precompute.sh` | Absorbed into ensure-fxcache step |
+| Modify | `infra/k8s/argo/kustomization.yaml` | Remove deleted templates, add new one |
+
+---
+
+### Task 1: Create the unified `train-template.yaml`
+
+**Files:**
+- Create: `infra/k8s/argo/train-template.yaml`
+
+- [ ] **Step 1: Write the template header, parameters, volumes, and DAG**
+
+```yaml
+# Unified training pipeline: ensure-binary → ensure-fxcache → train
+#
+# Smart caching:
+#   - Binaries cached on PVC at /data/bin/{sha}/ — skips compile if present
+#   - fxcache cached on feature-cache PVC — skips precompute if present
+#   - GPU warmup runs in parallel with ensure-binary
+#
+# Usage:
+#   ./scripts/argo-train.sh dqn
+#   ./scripts/argo-train.sh dqn --sha abc1234 --epochs 100
+---
+apiVersion: argoproj.io/v1alpha1
+kind: WorkflowTemplate
+metadata:
+  name: train
+  namespace: foxhunt
+  labels:
+    app.kubernetes.io/name: train
+    app.kubernetes.io/part-of: foxhunt
+spec:
+  entrypoint: pipeline
+  onExit: notify-result
+  serviceAccountName: argo-workflow
+  podMetadata:
+    labels:
+      app.kubernetes.io/part-of: foxhunt
+      app.kubernetes.io/component: train
+  securityContext:
+    fsGroup: 0
+  ttlStrategy:
+    secondsAfterCompletion: 3600
+  activeDeadlineSeconds: 28800
+
+  arguments:
+    parameters:
+      - name: commit-sha
+        value: HEAD
+      - name: git-branch
+        value: main
+      - name: cuda-compute-cap
+        value: "90"
+      - name: model
+        value: dqn
+      - name: gpu-pool
+        value: ci-training-h100
+      - name: hyperopt-trials
+        value: "20"
+      - name: hyperopt-epochs
+        value: "8"
+      - name: train-epochs
+        value: "50"
+      - name: symbol
+        value: ES.FUT
+      - name: initial-capital
+        value: "35000"
+      - name: tx-cost-bps
+        value: "0.1"
+      - name: tick-size
+        value: "0.25"
+      - name: spread-ticks
+        value: "1.0"
+      - name: min-hold-bars
+        value: "5"
+
+  volumes:
+    - name: git-ssh-key
+      secret:
+        secretName: argo-git-ssh-key
+        defaultMode: 256
+    - name: training-data
+      persistentVolumeClaim:
+        claimName: training-data-pvc
+    - name: cargo-target-cuda
+      persistentVolumeClaim:
+        claimName: cargo-target-cuda
+    - name: feature-cache
+      persistentVolumeClaim:
+        claimName: feature-cache-pvc
+
+  volumeClaimTemplates:
+    - metadata:
+        name: workspace
+      spec:
+        accessModes: ["ReadWriteOnce"]
+        storageClassName: scw-bssd
+        resources:
+          requests:
+            storage: 5Gi
+
+  templates:
+    - name: pipeline
+      dag:
+        tasks:
+          - name: ensure-binary
+            template: ensure-binary
+          - name: gpu-warmup
+            template: gpu-warmup
+          - name: ensure-fxcache
+            template: ensure-fxcache
+            dependencies: [ensure-binary]
+          - name: hyperopt
+            template: hyperopt
+            dependencies: [ensure-binary, ensure-fxcache, gpu-warmup]
+            when: "{{workflow.parameters.hyperopt-trials}} != 0"
+          - name: train-best
+            template: train-best
+            dependencies: [ensure-binary, ensure-fxcache, gpu-warmup, hyperopt]
+          - name: evaluate
+            template: evaluate
+            dependencies: [train-best]
+          - name: upload-results
+            template: upload-results
+            dependencies: [evaluate]
+```
+
+- [ ] **Step 2: Write the `ensure-binary` template**
+
+This is the key innovation — checks PVC cache first, compiles only on miss. Append to the same file:
+
+```yaml
+    # ── ensure-binary: cache-or-compile training binaries ──
+    # Checks /data/bin/{sha}/ on PVC. If binaries exist, exits immediately.
+    # If not, clones repo, builds, copies to /data/bin/{sha}/, cleans old SHAs.
+    - name: ensure-binary
+      nodeSelector:
+        k8s.scaleway.com/pool-name: ci-compile-cpu
+      tolerations:
+        - key: node.cilium.io/agent-not-ready
+          operator: Exists
+          effect: NoSchedule
+      container:
+        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
+        imagePullPolicy: Always
+        command: ["/bin/bash", "-c"]
+        env:
+          - name: SQLX_OFFLINE
+            value: "true"
+          - name: CARGO_TERM_COLOR
+            value: always
+          - name: CARGO_TARGET_DIR
+            value: /cargo-target
+          - name: CARGO_HOME
+            value: /cargo-target/cargo-home
+          - name: GITLAB_PAT
+            valueFrom:
+              secretKeyRef:
+                name: gitlab-pat
+                key: token
+        resources:
+          requests:
+            cpu: "14"
+            memory: 32Gi
+          limits:
+            cpu: "30"
+            memory: 64Gi
+        volumeMounts:
+          - name: git-ssh-key
+            mountPath: /etc/git-ssh
+            readOnly: true
+          - name: cargo-target-cuda
+            mountPath: /cargo-target
+          - name: training-data
+            mountPath: /data
+        args:
+          - |
+            set -e
+            BRANCH="{{workflow.parameters.git-branch}}"
+            REPO="ssh://git@gitlab-gitlab-shell.foxhunt.svc.cluster.local:2222/root/foxhunt.git"
+            BUILD="/cargo-target/src"
+
+            # Resolve commit SHA
+            mkdir -p ~/.ssh
+            cp /etc/git-ssh/ssh-privatekey ~/.ssh/id_ed25519
+            chmod 600 ~/.ssh/id_ed25519
+            printf 'Host *\n  StrictHostKeyChecking no\n  UserKnownHostsFile /dev/null\n' > ~/.ssh/config
+            chmod 600 ~/.ssh/config
+
+            git config --global --add safe.directory "$BUILD"
+
+            SHA="{{workflow.parameters.commit-sha}}"
+            if [ "$SHA" = "HEAD" ]; then
+              # Need to resolve HEAD to actual SHA
+              if [ -d "$BUILD/.git" ]; then
+                cd "$BUILD" && git fetch origin
+                SHA=$(git rev-parse "origin/$BRANCH")
+                cd /
+              else
+                git clone --filter=blob:none --depth=1 "$REPO" /tmp/sha-resolve
+                SHA=$(git -C /tmp/sha-resolve rev-parse HEAD)
+                rm -rf /tmp/sha-resolve
+              fi
+            fi
+            SHORT_SHA=$(echo "$SHA" | cut -c1-9)
+            echo "Resolved SHA: $SHORT_SHA"
+
+            # ── Cache check ──
+            BIN_DIR="/data/bin/$SHORT_SHA"
+            MODEL="{{workflow.parameters.model}}"
+            case "$MODEL" in
+              dqn|ppo) BINARIES="hyperopt_baseline_rl train_baseline_rl evaluate_baseline precompute_features" ;;
+              *)       BINARIES="hyperopt_baseline_supervised train_baseline_supervised evaluate_supervised precompute_features" ;;
+            esac
+
+            ALL_CACHED=true
+            for bin in $BINARIES; do
+              if [ ! -x "$BIN_DIR/$bin" ]; then
+                ALL_CACHED=false
+                break
+              fi
+            done
+
+            if $ALL_CACHED; then
+              echo "=== Binary cache HIT for $SHORT_SHA ==="
+              ls -lh "$BIN_DIR/"
+              echo "$SHORT_SHA" > /tmp/sha
+              exit 0
+            fi
+
+            echo "=== Binary cache MISS for $SHORT_SHA — compiling ==="
+
+            # Checkout
+            if [ -d "$BUILD/.git" ]; then
+              cd "$BUILD"
+              git fetch origin
+              CURRENT=$(git rev-parse HEAD 2>/dev/null || echo "none")
+              if [ "$CURRENT" != "$SHA" ]; then
+                git checkout --force "$SHA"
+                git clean -fd
+              fi
+            else
+              git clone --filter=blob:none "$REPO" "$BUILD"
+              cd "$BUILD"
+              git checkout "$SHA"
+            fi
+
+            export PATH="${CARGO_HOME}/bin:${PATH}"
+            export CUDA_COMPUTE_CAP={{workflow.parameters.cuda-compute-cap}}
+
+            # Prune build artifacts if PVC exceeds 25GB
+            TARGET_SIZE_MB=$(du -sm "$CARGO_TARGET_DIR" 2>/dev/null | cut -f1 || echo 0)
+            if [ "$TARGET_SIZE_MB" -gt 25000 ]; then
+              echo "PVC exceeds 25GB, pruning..."
+              rm -rf "$CARGO_TARGET_DIR/release" "$CARGO_TARGET_DIR/debug"
+            fi
+
+            ML_EXAMPLE_ARGS=""
+            for ex in $BINARIES; do
+              ML_EXAMPLE_ARGS="$ML_EXAMPLE_ARGS --example $ex"
+            done
+
+            echo "=== Building: $BINARIES ==="
+            cargo build --release -p ml --features ml/cuda $ML_EXAMPLE_ARGS
+
+            mkdir -p "$BIN_DIR"
+            for bin in $BINARIES; do
+              cp "$CARGO_TARGET_DIR/release/examples/$bin" "$BIN_DIR/"
+            done
+            strip "$BIN_DIR/"*
+
+            echo "=== Binaries cached at $BIN_DIR ==="
+            ls -lh "$BIN_DIR/"
+
+            # Cleanup: keep only last 5 SHA dirs
+            ls -dt /data/bin/*/ 2>/dev/null | tail -n +6 | xargs rm -rf 2>/dev/null || true
+
+            echo "$SHORT_SHA" > /tmp/sha
+      outputs:
+        parameters:
+          - name: sha
+            valueFrom:
+              path: /tmp/sha
+```
+
+- [ ] **Step 3: Write the `ensure-fxcache` template**
+
+```yaml
+    # ── ensure-fxcache: run precompute if cache miss ──
+    # The precompute binary computes its own cache key and skips if output exists.
+    - name: ensure-fxcache
+      nodeSelector:
+        k8s.scaleway.com/pool-name: ci-compile-cpu
+      tolerations:
+        - key: node.cilium.io/agent-not-ready
+          operator: Exists
+          effect: NoSchedule
+      container:
+        image: ubuntu:24.04
+        command: ["/bin/bash", "-c"]
+        env:
+          - name: RUST_LOG
+            value: info
+        resources:
+          requests:
+            cpu: "4"
+            memory: 16Gi
+          limits:
+            cpu: "28"
+            memory: 56Gi
+        volumeMounts:
+          - name: training-data
+            mountPath: /data
+            readOnly: true
+          - name: feature-cache
+            mountPath: /feature-cache
+        args:
+          - |
+            set -e
+            SHA="{{tasks.ensure-binary.outputs.parameters.sha}}"
+            BINARY="/data/bin/$SHA/precompute_features"
+
+            if [ ! -x "$BINARY" ]; then
+              echo "ERROR: $BINARY not found"
+              exit 1
+            fi
+
+            echo "=== Feature precompute (skip-if-exists) ==="
+            $BINARY \
+              --data-dir /data/futures-baseline \
+              --mbp10-data-dir /data/futures-baseline-mbp10 \
+              --trades-data-dir /data/futures-baseline-trades \
+              --output-dir /feature-cache \
+              --symbol {{workflow.parameters.symbol}} \
+              --yes
+```
+
+- [ ] **Step 4: Write the `gpu-warmup` template**
+
+```yaml
+    # ── gpu-warmup: trigger GPU node autoscale in parallel ──
+    - name: gpu-warmup
+      nodeSelector:
+        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
+      tolerations:
+        - key: nvidia.com/gpu
+          operator: Exists
+          effect: NoSchedule
+        - key: node.cilium.io/agent-not-ready
+          operator: Exists
+          effect: NoSchedule
+      container:
+        image: busybox:1.37
+        command: ["/bin/sh", "-c"]
+        args:
+          - |
+            echo "GPU warmup: node scheduled, autoscale triggered"
+        resources:
+          requests:
+            nvidia.com/gpu: "1"
+            cpu: 100m
+            memory: 64Mi
+          limits:
+            nvidia.com/gpu: "1"
+            cpu: 200m
+            memory: 128Mi
+```
+
+- [ ] **Step 5: Write `hyperopt`, `train-best`, `evaluate`, `upload-results`, `notify-result` templates**
+
+These are copied from `compile-and-train-template.yaml` with one change: binaries are read from `/data/bin/{sha}/` instead of `/workspace/bin/`. Replace all `export PATH="/workspace/bin:$PATH"` with:
+
+```bash
+SHA="{{tasks.ensure-binary.outputs.parameters.sha}}"
+export PATH="/data/bin/$SHA:$PATH"
+```
+
+And replace `feature-cache-dir` parameter references with the hardcoded `/feature-cache` path.
+
+The `hyperopt` template (lines 411-503 of old file), `train-best` (505-587), `evaluate` (589-654), `upload-results` (656-704), and `notify-result` (706-747) are otherwise identical — copy them with the PATH fix above.
+
+- [ ] **Step 6: Verify the template is valid YAML**
+
+Run: `python3 -c "import yaml; yaml.safe_load(open('infra/k8s/argo/train-template.yaml'))" && echo "Valid YAML"`
+Expected: `Valid YAML`
+
+- [ ] **Step 7: Commit**
+
+```bash
+git add infra/k8s/argo/train-template.yaml
+git commit -m "feat: unified train workflow template — cache-or-compile + cache-or-precompute"
+```
+
+---
+
+### Task 2: Rewrite `scripts/argo-train.sh`
+
+**Files:**
+- Modify: `scripts/argo-train.sh`
+
+- [ ] **Step 1: Replace the entire script**
+
+```bash
+#!/usr/bin/env bash
+# Train a model via Argo Workflows.
+#
+# Usage:
+#   ./scripts/argo-train.sh dqn                          # defaults: HEAD, H100, 50 epochs
+#   ./scripts/argo-train.sh dqn --sha abc1234            # specific commit
+#   ./scripts/argo-train.sh dqn --epochs 100 --trials 40 # override training params
+#   ./scripts/argo-train.sh ppo --gpu-pool ci-training    # L40S instead of H100
+#   ./scripts/argo-train.sh dqn --baseline                # skip hyperopt
+#   ./scripts/argo-train.sh dqn --watch                  # follow logs
+#
+# Supported models:
+#   RL:          dqn, ppo
+#   Supervised:  tft, mamba2, tggn, tlob, liquid, kan, xlstm, diffusion
+#
+# Requires: argo CLI
+set -euo pipefail
+
+SHA="HEAD"
+BRANCH="main"
+TRIALS=""
+EPOCHS=""
+GPU_POOL=""
+SYMBOL=""
+WATCH=false
+BASELINE=false
+CAPITAL=""
+
+usage() {
+  cat <<EOF
+Usage: $(basename "$0") <model> [OPTIONS]
+
+Models:
+  dqn, ppo                                          (RL)
+  tft, mamba2, tggn, tlob, liquid, kan, xlstm       (supervised)
+
+Options:
+  --sha <commit>       Git commit SHA (default: HEAD)
+  --branch <branch>    Git branch (default: main)
+  --trials <n>         Hyperopt trials (default: 20)
+  --epochs <n>         Training epochs (default: 50)
+  --gpu-pool <pool>    GPU node pool (default: ci-training-h100)
+  --symbol <sym>       Trading symbol (default: ES.FUT)
+  --capital <n>        Initial capital (default: 35000)
+  --baseline           Skip hyperopt (trials=0)
+  --watch              Follow workflow logs
+  -h, --help           Show this help
+EOF
+  exit 0
+}
+
+[[ $# -eq 0 ]] && { echo "Error: model argument required"; usage; }
+
+MODEL="$1"; shift
+
+case "$MODEL" in
+  dqn|ppo|tft|mamba2|tggn|tlob|liquid|kan|xlstm|diffusion) ;;
+  *) echo "Error: unknown model '$MODEL'"; usage ;;
+esac
+
+while [[ $# -gt 0 ]]; do
+  case $1 in
+    --sha)       SHA="$2"; shift 2 ;;
+    --branch)    BRANCH="$2"; shift 2 ;;
+    --trials)    TRIALS="$2"; shift 2 ;;
+    --epochs)    EPOCHS="$2"; shift 2 ;;
+    --gpu-pool)  GPU_POOL="$2"; shift 2 ;;
+    --symbol)    SYMBOL="$2"; shift 2 ;;
+    --capital)   CAPITAL="$2"; shift 2 ;;
+    --baseline)  BASELINE=true; shift ;;
+    --watch)     WATCH=true; shift ;;
+    -h|--help)   usage ;;
+    *)           echo "Unknown option: $1"; usage ;;
+  esac
+done
+
+CMD="argo submit -n foxhunt --from=wftmpl/train"
+CMD="$CMD -p commit-sha=$SHA"
+CMD="$CMD -p git-branch=$BRANCH"
+CMD="$CMD -p model=$MODEL"
+
+[[ -n "$TRIALS" ]]    && CMD="$CMD -p hyperopt-trials=$TRIALS"
+[[ -n "$EPOCHS" ]]    && CMD="$CMD -p train-epochs=$EPOCHS"
+[[ -n "$GPU_POOL" ]]  && CMD="$CMD -p gpu-pool=$GPU_POOL"
+[[ -n "$SYMBOL" ]]    && CMD="$CMD -p symbol=$SYMBOL"
+[[ -n "$CAPITAL" ]]   && CMD="$CMD -p initial-capital=$CAPITAL"
+
+$BASELINE && CMD="$CMD -p hyperopt-trials=0"
+$WATCH    && CMD="$CMD --watch"
+
+echo "Submitting $MODEL training workflow..."
+echo "  sha:    $SHA"
+echo "  branch: $BRANCH"
+echo "  model:  $MODEL"
+$BASELINE              && echo "  mode:   baseline (no hyperopt)"
+[[ -n "$TRIALS" ]]     && echo "  trials: $TRIALS"
+[[ -n "$EPOCHS" ]]     && echo "  epochs: $EPOCHS"
+[[ -n "$GPU_POOL" ]]   && echo "  gpu:    $GPU_POOL"
+echo ""
+eval "$CMD"
+```
+
+- [ ] **Step 2: Verify script is executable**
+
+Run: `bash -n scripts/argo-train.sh && echo "Syntax OK"`
+Expected: `Syntax OK`
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add scripts/argo-train.sh
+git commit -m "feat: rewrite argo-train.sh — single template, --sha for commit pinning"
+```
+
+---
+
+### Task 3: Delete dead templates and scripts
+
+**Files:**
+- Delete: `infra/k8s/argo/compile-and-train-template.yaml`
+- Delete: `infra/k8s/argo/train-dqn-template.yaml`
+- Delete: `infra/k8s/argo/train-baseline-rl-template.yaml`
+- Delete: `infra/k8s/argo/train-supervised-template.yaml`
+- Delete: `infra/k8s/argo/training-workflow-template.yaml`
+- Delete: `infra/k8s/argo/precompute-features-template.yaml`
+- Delete: `infra/k8s/argo/train-ppo-template.yaml`
+- Delete: `scripts/argo-precompute.sh`
+- Modify: `infra/k8s/argo/kustomization.yaml`
+
+- [ ] **Step 1: Delete the template files**
+
+```bash
+rm infra/k8s/argo/compile-and-train-template.yaml
+rm infra/k8s/argo/train-dqn-template.yaml
+rm infra/k8s/argo/train-baseline-rl-template.yaml
+rm infra/k8s/argo/train-supervised-template.yaml
+rm infra/k8s/argo/training-workflow-template.yaml
+rm infra/k8s/argo/precompute-features-template.yaml
+rm infra/k8s/argo/train-ppo-template.yaml
+rm scripts/argo-precompute.sh
+```
+
+- [ ] **Step 2: Update kustomization.yaml**
+
+Read `infra/k8s/argo/kustomization.yaml`, remove references to deleted templates, add `train-template.yaml`.
+
+- [ ] **Step 3: Commit**
+
+```bash
+git add -A infra/k8s/argo/ scripts/
+git commit -m "cleanup: delete 7 dead training templates + precompute script, update kustomization"
+```
+
+---
+
+### Task 4: Deploy to cluster and delete old templates
+
+**Files:**
+- No file changes — cluster operations only
+
+- [ ] **Step 1: Apply the new template to the cluster**
+
+```bash
+kubectl apply -f infra/k8s/argo/train-template.yaml
+```
+
+Expected: `workflowtemplate.argoproj.io/train created`
+
+- [ ] **Step 2: Delete old templates from the cluster**
+
+```bash
+kubectl delete wftmpl -n foxhunt compile-and-train train-dqn train-baseline-rl train-supervised training-pipeline precompute-features train-ppo 2>/dev/null || true
+```
+
+- [ ] **Step 3: Verify only the new template exists (plus the non-training ones)**
+
+```bash
+kubectl get wftmpl -n foxhunt
+```
+
+Expected: `train` plus `compile-and-deploy`, `databento-download`, `gpu-test-pipeline`, `build-ci-image`, `ci-pipeline`
+
+- [ ] **Step 4: Commit (no file changes — just verify)**
+
+No commit needed. Cluster state is updated.
+
+---
+
+### Task 5: Test the new workflow — baseline DQN on H100
+
+**Files:**
+- No file changes — integration test
+
+- [ ] **Step 1: Run a baseline DQN training (no hyperopt, 5 epochs)**
+
+```bash
+./scripts/argo-train.sh dqn --baseline --epochs 5 --watch
+```
+
+Expected: 
+- `ensure-binary`: cache HIT (binary was compiled by the terminated workflow earlier) OR cache MISS → compile ~10 min
+- `ensure-fxcache`: cache HIT (fxcache was generated earlier) OR run precompute ~4 min
+- `gpu-warmup`: completes in ~3 min (node autoscale)
+- `train-best`: runs 5 epochs on H100 with TF32 tensor cores
+- Watch for epoch timing — should be ~10s/epoch on H100 (was 77s with old CUDA core path)
+
+- [ ] **Step 2: Verify epoch timing in logs**
+
+```bash
+argo logs <workflow-name> -n foxhunt | grep 'duration_secs\|duration='
+```
+
+Expected: epoch times significantly faster than 77s (the old bf16/CUDA core baseline).
+
+- [ ] **Step 3: Push all changes**
+
+```bash
+git push origin main
+```
diff --git a/infra/k8s/argo/compile-and-train-template.yaml b/infra/k8s/argo/compile-and-train-template.yaml
deleted file mode 100644
index ef4f17923..000000000
--- a/infra/k8s/argo/compile-and-train-template.yaml
+++ /dev/null
@@ -1,746 +0,0 @@
-# Compile-and-Train: unified workflow for code change → training run.
-#
-# DAG:
-#   compile-training ──┐
-#   gpu-warmup ────────┼──> fetch-binary ──> hyperopt ──> train-best ──> evaluate ──> upload-results
-#
-# Image builds handled by ci-pipeline, not here.
-# Compile and GPU warmup run in parallel (~5.5 min compile, ~3-5 min autoscale).
-# When both finish, fetch-binary downloads the freshly compiled binaries and
-# training starts immediately — no manual CI wait or package seeding.
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: compile-and-train
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: compile-and-train
-    app.kubernetes.io/part-of: foxhunt
-    app.kubernetes.io/component: gpu-test
-spec:
-  entrypoint: pipeline
-  onExit: notify-result
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: compile-and-train
-  # No runAsUser — compile (ci-builder) needs root, training runtime defaults
-  # to its image user. fsGroup ensures workspace PVC is writable by all pods.
-  securityContext:
-    fsGroup: 0
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 28800  # 8 hours (compile + full training)
-
-  arguments:
-    parameters:
-      # ── Compile ──
-      - name: commit-sha
-        value: HEAD
-      - name: git-branch
-        value: main
-      - name: cuda-compute-cap
-        value: "90"
-      # ── Training ──
-      - name: model
-        value: dqn
-      - name: gpu-pool
-        value: ci-training-h100
-      - name: hyperopt-trials
-        value: "20"
-      - name: hyperopt-epochs
-        value: "8"
-      - name: train-epochs
-        value: "50"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: feature-cache-dir
-        value: /feature-cache
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: min-hold-bars
-        value: "5"
-      - name: initial-capital
-        value: "35000"
-      - name: ensemble-top-k
-        value: "5"
-
-  volumes:
-    - name: git-ssh-key
-      secret:
-        secretName: argo-git-ssh-key
-        defaultMode: 256
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-    - name: cargo-target-cuda
-      persistentVolumeClaim:
-        claimName: cargo-target-cuda
-    - name: feature-cache
-      persistentVolumeClaim:
-        claimName: feature-cache-pvc
-
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: pipeline
-      dag:
-        tasks:
-          - name: compile-training
-            template: compile-training
-          - name: gpu-warmup
-            template: gpu-warmup
-          - name: fetch-binary
-            template: fetch-binary
-            dependencies: [compile-training]
-            arguments:
-              parameters:
-                - name: tag
-                  value: "{{tasks.compile-training.outputs.parameters.tag}}"
-          - name: precompute
-            template: precompute
-            dependencies: [fetch-binary]
-          - name: hyperopt
-            template: hyperopt
-            dependencies: [fetch-binary, gpu-warmup, precompute]
-            when: "{{workflow.parameters.hyperopt-trials}} != 0"
-          - name: train-best
-            template: train-best
-            dependencies: [fetch-binary, gpu-warmup, hyperopt]
-          - name: evaluate
-            template: evaluate
-            dependencies: [train-best]
-          - name: upload-results
-            template: upload-results
-            dependencies: [evaluate]
-
-    # ── compile-training: build from source on ci-compile-cpu ──
-    - name: compile-training
-      outputs:
-        parameters:
-          - name: tag
-            valueFrom:
-              path: /tmp/tag
-      nodeSelector:
-        k8s.scaleway.com/pool-name: ci-compile-cpu
-      tolerations:
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: Always
-        command: ["/bin/sh", "-c"]
-        env:
-          - name: SQLX_OFFLINE
-            value: "true"
-          - name: CARGO_TERM_COLOR
-            value: always
-          - name: CARGO_TARGET_DIR
-            value: /cargo-target
-          - name: CARGO_HOME
-            value: /cargo-target/cargo-home
-          - name: GITLAB_PAT
-            valueFrom:
-              secretKeyRef:
-                name: gitlab-pat
-                key: token
-        resources:
-          requests:
-            cpu: "14"
-            memory: 32Gi
-          limits:
-            cpu: "30"
-            memory: 64Gi
-        volumeMounts:
-          - name: git-ssh-key
-            mountPath: /etc/git-ssh
-            readOnly: true
-          - name: cargo-target-cuda
-            mountPath: /cargo-target
-        args:
-          - |
-            set -e
-            SHA="{{workflow.parameters.commit-sha}}"
-            BRANCH="{{workflow.parameters.git-branch}}"
-
-            # SSH setup
-            mkdir -p ~/.ssh
-            cp /etc/git-ssh/ssh-privatekey ~/.ssh/id_ed25519
-            chmod 600 ~/.ssh/id_ed25519
-            printf 'Host *\n  StrictHostKeyChecking no\n  UserKnownHostsFile /dev/null\n' > ~/.ssh/config
-            chmod 600 ~/.ssh/config
-
-            REPO="ssh://git@gitlab-gitlab-shell.foxhunt.svc.cluster.local:2222/root/foxhunt.git"
-            BUILD="/cargo-target/src"
-
-            # PVC may be owned by a different UID from a previous run
-            git config --global --add safe.directory "$BUILD"
-
-            # Persistent checkout on PVC — only changed files get new mtimes,
-            # so cargo skips recompiling unchanged workspace crates.
-            if [ -d "$BUILD/.git" ]; then
-              cd "$BUILD"
-              git fetch origin
-              # Resolve SHA: "HEAD" means remote branch tip, not local HEAD
-              if [ "$SHA" = "HEAD" ]; then
-                SHA=$(git rev-parse "origin/$BRANCH")
-              fi
-              CURRENT=$(git rev-parse HEAD 2>/dev/null || echo "none")
-              if [ "$CURRENT" = "$SHA" ]; then
-                echo "=== Already at $SHA, skipping checkout ==="
-              else
-                echo "=== Updating checkout: $(echo $CURRENT | cut -c1-8) -> $(echo $SHA | cut -c1-8) ==="
-                git checkout --force "$SHA"
-                git clean -fd
-              fi
-            else
-              echo "=== Initial clone (first run) ==="
-              git clone --filter=blob:none "$REPO" "$BUILD"
-              cd "$BUILD"
-              if [ "$SHA" = "HEAD" ]; then
-                git checkout "origin/$BRANCH"
-              else
-                git checkout "$SHA"
-              fi
-            fi
-            SHORT_SHA=$(git rev-parse --short HEAD)
-            echo "Checked out ${SHORT_SHA}"
-
-            # Ensure cargo home registry is on PVC
-            export PATH="${CARGO_HOME}/bin:${PATH}"
-
-            export CUDA_COMPUTE_CAP={{workflow.parameters.cuda-compute-cap}}
-
-            # Prune build artifacts if PVC exceeds 25GB (prevents unbounded growth)
-            TARGET_SIZE_MB=$(du -sm "$CARGO_TARGET_DIR" 2>/dev/null | cut -f1 || echo 0)
-            echo "PVC usage: ${TARGET_SIZE_MB}MB"
-            if [ "$TARGET_SIZE_MB" -gt 25000 ]; then
-              echo "PVC exceeds 25GB limit, pruning build artifacts..."
-              rm -rf "$CARGO_TARGET_DIR/release" "$CARGO_TARGET_DIR/debug"
-            fi
-
-            # Derive needed binaries from model parameter (only build what's used)
-            MODEL="{{workflow.parameters.model}}"
-            case "$MODEL" in
-              dqn|ppo)
-                EXAMPLES="hyperopt_baseline_rl train_baseline_rl evaluate_baseline precompute_features"
-                ;;
-              *)
-                EXAMPLES="hyperopt_baseline_supervised train_baseline_supervised evaluate_supervised"
-                ;;
-            esac
-
-            ML_EXAMPLE_ARGS=""
-            for ex in $EXAMPLES; do
-              ML_EXAMPLE_ARGS="$ML_EXAMPLE_ARGS --example $ex"
-            done
-
-            echo "=== Building training binaries for $MODEL: $EXAMPLES (incremental) ==="
-            echo "    CUDA_COMPUTE_CAP=$CUDA_COMPUTE_CAP, target dir on PVC /cargo-target"
-            cargo build --release -p ml --features ml/cuda $ML_EXAMPLE_ARGS
-
-            mkdir -p "$BUILD/bin/training"
-            for bin in $EXAMPLES; do
-              cp "$CARGO_TARGET_DIR/release/examples/$bin" "$BUILD/bin/training/"
-            done
-            strip "$BUILD/bin/training/"*
-
-            echo "=== Training binaries ==="
-            ls -lh "$BUILD/bin/training/"
-
-            # Upload to GitLab packages under commit SHA
-            GITLAB="http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181"
-            TAG="dev-${SHORT_SHA}"
-            echo "=== Uploading as ${TAG} ==="
-            for bin in "$BUILD/bin/training/"*; do
-              BIN_NAME=$(basename "$bin")
-              curl -f --upload-file "$bin" \
-                -H "PRIVATE-TOKEN: ${GITLAB_PAT}" \
-                "${GITLAB}/api/v4/projects/1/packages/generic/foxhunt-training/${TAG}/${BIN_NAME}"
-            done
-
-            echo "=== Compile + upload done (${TAG}) ==="
-            echo "$TAG" > /tmp/tag
-
-    # ── precompute: generate fxcache if data changed (skips if cache exists) ──
-    # Pure CPU work — no GPU needed. Runs on compile node (same as compile-training)
-    # to reuse workspace PVC attachment.
-    - name: precompute
-      nodeSelector:
-        k8s.scaleway.com/pool-name: ci-compile-cpu
-      tolerations:
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: ubuntu:24.04
-        command: ["/bin/bash", "-c"]
-        env:
-          - name: RUST_LOG
-            value: info
-        args:
-          - |
-            set -e
-            BINARY=/workspace/bin/precompute_features
-            if [ ! -f "$BINARY" ]; then
-              echo "ERROR: precompute_features not found in /workspace/bin/"
-              exit 1
-            fi
-            $BINARY \
-              --data-dir /data/futures-baseline \
-              --mbp10-data-dir /data/futures-baseline-mbp10 \
-              --trades-data-dir /data/futures-baseline-trades \
-              --output-dir /feature-cache \
-              --symbol ES.FUT \
-              --yes
-        volumeMounts:
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-          - name: feature-cache
-            mountPath: /feature-cache
-          - name: workspace
-            mountPath: /workspace
-        resources:
-          requests:
-            cpu: "4"
-            memory: 16Gi
-          limits:
-            cpu: "28"
-            memory: 56Gi
-
-    # ── gpu-warmup: trigger GPU autoscale during compile ──
-    - name: gpu-warmup
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: busybox:1.37
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            echo "GPU warmup: triggering node autoscale..."
-            echo "GPU node scheduled, exiting to free resources"
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: 100m
-            memory: 64Mi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: 200m
-            memory: 128Mi
-
-    # ── fetch-binary: download freshly compiled binaries to workspace ──
-    - name: fetch-binary
-      inputs:
-        parameters:
-          - name: tag
-      nodeSelector:
-        k8s.scaleway.com/pool-name: platform
-      container:
-        image: curlimages/curl:8.12.1
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            mkdir -p /workspace/bin
-            GITLAB_API="http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181/api/v4"
-
-            TAG="{{inputs.parameters.tag}}"
-            echo "Fetching training binaries from GitLab package: foxhunt-training/${TAG}"
-
-            BINARIES="hyperopt_baseline_rl hyperopt_baseline_supervised train_baseline_rl train_baseline_supervised evaluate_baseline evaluate_supervised precompute_features"
-            for bin in $BINARIES; do
-              echo "  Downloading ${bin}..."
-              curl -fSL -o "/workspace/bin/${bin}" \
-                -H "PRIVATE-TOKEN: ${GITLAB_PAT}" \
-                "${GITLAB_API}/projects/1/packages/generic/foxhunt-training/${TAG}/${bin}" || {
-                echo "WARN: ${bin} not found, skipping"
-                continue
-              }
-            done
-
-            chmod +x /workspace/bin/*
-            echo "Fetched binaries:"
-            ls -lh /workspace/bin/
-        env:
-          - name: GITLAB_PAT
-            valueFrom:
-              secretKeyRef:
-                name: gitlab-pat
-                key: token
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-        resources:
-          requests:
-            cpu: 100m
-            memory: 128Mi
-          limits:
-            cpu: 500m
-            memory: 256Mi
-
-    # ── hyperopt: PSO/TPE optimization on GPU ──
-    - name: hyperopt
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-          - name: CUBLAS_WORKSPACE_CONFIG
-            value: ":4096:8"
-          - name: CARGO_TARGET_DIR
-            value: /workspace
-          - name: FOXHUNT_FEATURE_CACHE_DIR
-            value: "{{workflow.parameters.feature-cache-dir}}"
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            # Remove CUDA stubs from LD_LIBRARY_PATH — stubs shadow real NVIDIA drivers
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-            nvidia-smi
-
-            # Training data available via training-data-pvc mounted at /data
-            echo "=== Training data (PVC) ==="
-            echo "OHLCV: $(ls {{workflow.parameters.data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-            echo "MBP-10: $(ls {{workflow.parameters.mbp10-data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-            echo "Trades: $(ls {{workflow.parameters.trades-data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-
-            mkdir -p /workspace/output/hyperopt
-
-            echo "=== Feature cache: $(ls ${FOXHUNT_FEATURE_CACHE_DIR}/*.fxcache 2>/dev/null | wc -l) fxcache files ==="
-
-            MODEL="{{workflow.parameters.model}}"
-            case "$MODEL" in
-              dqn|ppo) BINARY=hyperopt_baseline_rl ;;
-              *)       BINARY=hyperopt_baseline_supervised ;;
-            esac
-
-            echo "=== Running hyperopt: $MODEL (${BINARY}) ==="
-            echo "  Trials: {{workflow.parameters.hyperopt-trials}}, Epochs: {{workflow.parameters.hyperopt-epochs}}"
-
-            # Single-pass 14D hyperopt: family-based search covers all dims at once.
-            # Fast phase searches all 14 dimensions (3 breakouts + 11 family intensities).
-            # Phased search was needed for the old 30D space; 14D is within PSO's efficient range.
-            ${BINARY} \
-              --model "$MODEL" \
-              --phase fast \
-              --trials {{workflow.parameters.hyperopt-trials}} \
-              --epochs {{workflow.parameters.hyperopt-epochs}} \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              --initial-capital {{workflow.parameters.initial-capital}} \
-              --min-hold-bars {{workflow.parameters.min-hold-bars}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --mbp10-data-dir {{workflow.parameters.mbp10-data-dir}} \
-              --trades-data-dir {{workflow.parameters.trades-data-dir}} \
-              --base-dir /workspace/output/hyperopt \
-              --output /workspace/output/${MODEL}_hyperopt_results.json
-
-            echo "=== Hyperopt complete (14D single-pass) ==="
-            cat /workspace/output/${MODEL}_hyperopt_results.json 2>/dev/null || echo "No results file"
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "4"
-            memory: 32Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "8"
-            memory: 64Gi
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-          - name: feature-cache
-            mountPath: /feature-cache
-            readOnly: true
-
-    # ── train-best: full training with best hyperparams ──
-    - name: train-best
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-          - name: CUBLAS_WORKSPACE_CONFIG
-            value: ":4096:8"
-          - name: FOXHUNT_FEATURE_CACHE_DIR
-            value: "{{workflow.parameters.feature-cache-dir}}"
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-            nvidia-smi
-
-            MODEL="{{workflow.parameters.model}}"
-
-            # Determine binary and hyperopt flag
-            case "$MODEL" in
-              dqn|ppo) BINARY=train_baseline_rl ;;
-              *)       BINARY=train_baseline_supervised ;;
-            esac
-
-            HYPEROPT_FLAG=""
-            if [ -f "/workspace/output/${MODEL}_hyperopt_results.json" ]; then
-              HYPEROPT_FLAG="--hyperopt-params /workspace/output/${MODEL}_hyperopt_results.json"
-              echo "  Using hyperopt results: ${MODEL}_hyperopt_results.json"
-            else
-              echo "  No hyperopt results — training with default hyperparams"
-            fi
-
-            echo "=== Training best: $MODEL ({{workflow.parameters.train-epochs}} epochs) ==="
-
-            stdbuf -oL ${BINARY} \
-              --model "$MODEL" \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              --initial-capital {{workflow.parameters.initial-capital}} \
-              --min-hold-bars {{workflow.parameters.min-hold-bars}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --mbp10-data-dir {{workflow.parameters.mbp10-data-dir}} \
-              --trades-data-dir {{workflow.parameters.trades-data-dir}} \
-              --output-dir /workspace/output \
-              --epochs {{workflow.parameters.train-epochs}} \
-              $HYPEROPT_FLAG
-
-            echo "=== Training complete ==="
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "4"
-            memory: 32Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "8"
-            memory: 64Gi
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-          - name: feature-cache
-            mountPath: /feature-cache
-            readOnly: true
-
-    # ── evaluate: run evaluation on trained model ──
-    - name: evaluate
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-
-            MODEL="{{workflow.parameters.model}}"
-            case "$MODEL" in
-              dqn|ppo) BINARY=evaluate_baseline ;;
-              *)       BINARY=evaluate_supervised ;;
-            esac
-
-            echo "=== Evaluating: $MODEL ==="
-            ${BINARY} \
-              --model "$MODEL" \
-              --symbol {{workflow.parameters.symbol}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --checkpoint-dir /workspace/output \
-              --output-dir /workspace/output/eval \
-              --initial-capital {{workflow.parameters.initial-capital}} \
-              --min-hold-bars {{workflow.parameters.min-hold-bars}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} || {
-              echo "WARN: Evaluation failed, continuing to upload available results"
-            }
-
-            echo "=== Evaluation complete ==="
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "2"
-            memory: 16Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "4"
-            memory: 32Gi
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-          - name: feature-cache
-            mountPath: /feature-cache
-            readOnly: true
-
-    # ── upload-results: push artifacts to GitLab packages ──
-    - name: upload-results
-      nodeSelector:
-        k8s.scaleway.com/pool-name: platform
-      container:
-        image: curlimages/curl:8.12.1
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            MODEL="{{workflow.parameters.model}}"
-            SYMBOL="{{workflow.parameters.symbol}}"
-            TIMESTAMP=$(date +%Y%m%d-%H%M%S)
-            GITLAB_API="http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181/api/v4"
-            PKG_NAME="foxhunt-training-results"
-            PKG_VERSION="${MODEL}-${SYMBOL}-${TIMESTAMP}"
-
-            echo "Uploading training artifacts to GitLab packages: ${PKG_NAME}/${PKG_VERSION}"
-
-            UPLOADED=0
-            find /workspace/output -type f | while read -r file; do
-              REL_PATH="${file#/workspace/output/}"
-              SAFE_NAME=$(echo "$REL_PATH" | tr '/' '--')
-              echo "  Uploading ${REL_PATH} as ${SAFE_NAME}..."
-              curl -f --upload-file "$file" \
-                -H "PRIVATE-TOKEN: ${GITLAB_PAT}" \
-                "${GITLAB_API}/projects/1/packages/generic/${PKG_NAME}/${PKG_VERSION}/${SAFE_NAME}" && \
-                UPLOADED=$((UPLOADED + 1)) || \
-                echo "  WARN: Failed to upload ${REL_PATH}"
-            done
-
-            echo "=== Upload complete (${UPLOADED} files) ==="
-            echo "Package: ${PKG_NAME}/${PKG_VERSION}"
-        env:
-          - name: GITLAB_PAT
-            valueFrom:
-              secretKeyRef:
-                name: gitlab-pat
-                key: token
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-        resources:
-          requests:
-            cpu: 100m
-            memory: 128Mi
-          limits:
-            cpu: 500m
-            memory: 256Mi
-
-    # ── notify-result: post workflow outcome to Mattermost ──
-    - name: notify-result
-      nodeSelector:
-        k8s.scaleway.com/pool-name: platform
-      container:
-        image: curlimages/curl:8.12.1
-        command: ["/bin/sh", "-c"]
-        env:
-          - name: WEBHOOK_URL
-            valueFrom:
-              secretKeyRef:
-                name: notification-webhook
-                key: webhook-url
-                optional: true
-        resources:
-          requests:
-            cpu: 50m
-            memory: 32Mi
-          limits:
-            cpu: 100m
-            memory: 64Mi
-        args:
-          - |
-            STATUS="{{workflow.status}}"
-            NAME="{{workflow.name}}"
-
-            if [ -z "$WEBHOOK_URL" ] || echo "$WEBHOOK_URL" | grep -q "PLACEHOLDER"; then
-              echo "No webhook configured, skipping notification"
-              exit 0
-            fi
-
-            if [ "$STATUS" = "Succeeded" ]; then
-              EMOJI=":white_check_mark:"
-            else
-              EMOJI=":x:"
-            fi
-
-            PAYLOAD="{\"username\":\"Argo CI\",\"text\":\"${EMOJI} **${NAME}** — ${STATUS} ({{workflow.duration}}s)\"}"
-
-            curl -sf -X POST -H 'Content-Type: application/json' \
-              -d "$PAYLOAD" "$WEBHOOK_URL" || echo "WARN: webhook post failed"
diff --git a/infra/k8s/argo/kustomization.yaml b/infra/k8s/argo/kustomization.yaml
index 49ecfbf34..57eb8269c 100644
--- a/infra/k8s/argo/kustomization.yaml
+++ b/infra/k8s/argo/kustomization.yaml
@@ -4,12 +4,8 @@ resources:
   - sccache-pvcs.yaml
   - ci-pipeline-template.yaml
   - compile-and-deploy-template.yaml
-  - compile-and-train-template.yaml
+  - train-template.yaml
   - build-ci-image-template.yaml
-  - training-workflow-template.yaml
-  - train-dqn-template.yaml
-  - train-ppo-template.yaml
-  - train-supervised-template.yaml
   - argo-workflow-netpol.yaml
   - ci-deploy-rbac.yaml
   - archive-rbac.yaml
diff --git a/infra/k8s/argo/precompute-features-template.yaml b/infra/k8s/argo/precompute-features-template.yaml
deleted file mode 100644
index b225d32bb..000000000
--- a/infra/k8s/argo/precompute-features-template.yaml
+++ /dev/null
@@ -1,122 +0,0 @@
-# Pre-compute DQN training features to .fxcache format (f32).
-#
-# Runs after data downloads complete. Reads DBN files from training-data PVC,
-# extracts all features (OHLCV 42-dim + targets 4-dim + OFI 8-dim), writes
-# flat binary cache for zero-overhead GPU loading.
-#
-# Usage:
-#   argo submit -n foxhunt --from=wftmpl/precompute-features
-#   ./scripts/argo-precompute.sh ES.FUT
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: precompute-features
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: precompute-features
-    app.kubernetes.io/part-of: foxhunt
-spec:
-  entrypoint: precompute
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: precompute-features
-  securityContext:
-    fsGroup: 1000
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 7200
-
-  arguments:
-    parameters:
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-dir
-        value: /data/futures-baseline-trades
-      - name: output-dir
-        value: /feature-cache
-      - name: node-pool
-        value: ci-compile-cpu
-
-  volumes:
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-    - name: feature-cache
-      persistentVolumeClaim:
-        claimName: feature-cache-pvc
-
-  templates:
-    - name: precompute
-      inputs:
-        parameters:
-          - name: symbol
-          - name: data-dir
-          - name: mbp10-dir
-          - name: trades-dir
-          - name: output-dir
-          - name: node-pool
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{inputs.parameters.node-pool}}"
-      container:
-        image: ubuntu:24.04
-        command: ["/bin/bash", "-c"]
-        env:
-          - name: RUST_LOG
-            value: info
-        resources:
-          requests:
-            cpu: "4"
-            memory: 16Gi
-          limits:
-            cpu: "28"
-            memory: 56Gi
-        volumeMounts:
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-          - name: feature-cache
-            mountPath: /feature-cache
-        args:
-          - |
-            set -e
-            # Binary deployed to PVC by compile-and-train workflow
-            BINARY=/data/bin/precompute_features
-            if [ ! -x "$BINARY" ]; then
-              echo "ERROR: $BINARY not found or not executable. Run compile-and-train first."
-              exit 1
-            fi
-
-            SYMBOL="{{inputs.parameters.symbol}}"
-            DATA_DIR="{{inputs.parameters.data-dir}}"
-            MBP10_DIR="{{inputs.parameters.mbp10-dir}}"
-            TRADES_DIR="{{inputs.parameters.trades-dir}}"
-            OUTPUT_DIR="{{inputs.parameters.output-dir}}"
-
-            echo "=== Feature Pre-Compute (f32) ==="
-            echo "Symbol:  $SYMBOL"
-            echo "OHLCV:   $DATA_DIR"
-            echo "MBP-10:  $MBP10_DIR"
-            echo "Trades:  $TRADES_DIR"
-            echo "Output:  $OUTPUT_DIR"
-            echo "Node:    $(hostname), CPUs: $(nproc), RAM: $(free -g | awk '/Mem:/{print $2}')G"
-            echo ""
-
-            $BINARY \
-              --data-dir "$DATA_DIR" \
-              --mbp10-data-dir "$MBP10_DIR" \
-              --trades-data-dir "$TRADES_DIR" \
-              --output-dir "$OUTPUT_DIR" \
-              --symbol "$SYMBOL" \
-              --yes
-
-            echo ""
-            echo "=== Output ==="
-            find "$OUTPUT_DIR" -name '*.fxcache' -exec ls -lh {} \;
-            du -sh "$OUTPUT_DIR"
diff --git a/infra/k8s/argo/train-baseline-rl-template.yaml b/infra/k8s/argo/train-baseline-rl-template.yaml
deleted file mode 100644
index 62f39eafc..000000000
--- a/infra/k8s/argo/train-baseline-rl-template.yaml
+++ /dev/null
@@ -1,102 +0,0 @@
-# Baseline RL training — compile-and-train with hyperopt disabled.
-#
-# Identical to train-dqn but hardcodes hyperopt-trials=0.
-# Compiles from source, skips hyperopt, trains with default params.
-#
-# Usage:
-#   argo submit -n foxhunt --from=wftmpl/train-baseline-rl
-#   argo submit -n foxhunt --from=wftmpl/train-baseline-rl -p train-epochs=100
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: train-baseline-rl
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: train-baseline-rl
-    app.kubernetes.io/part-of: foxhunt
-spec:
-  entrypoint: pipeline
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: compile-and-train
-  securityContext:
-    fsGroup: 0
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 14400
-
-  # Mirror ALL compile-and-train parameters — Argo resolves
-  # {{workflow.parameters.*}} in the caller's scope.
-  arguments:
-    parameters:
-      - name: commit-sha
-        value: HEAD
-      - name: git-branch
-        value: main
-      - name: cuda-compute-cap
-        value: "90"
-      - name: model
-        value: dqn
-      - name: gpu-pool
-        value: ci-training-h100
-      - name: hyperopt-trials
-        value: "0"
-      - name: hyperopt-epochs
-        value: "0"
-      - name: train-epochs
-        value: "50"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: feature-cache-dir
-        value: /data/feature-cache
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: min-hold-bars
-        value: "5"
-      - name: initial-capital
-        value: "35000"
-      - name: ensemble-top-k
-        value: "5"
-
-  volumes:
-    - name: git-ssh-key
-      secret:
-        secretName: argo-git-ssh-key
-        defaultMode: 256
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-    - name: cargo-target-cuda
-      persistentVolumeClaim:
-        claimName: cargo-target-cuda
-
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: pipeline
-      steps:
-        - - name: run
-            templateRef:
-              name: compile-and-train
-              template: pipeline
diff --git a/infra/k8s/argo/train-dqn-template.yaml b/infra/k8s/argo/train-dqn-template.yaml
deleted file mode 100644
index 297c2947f..000000000
--- a/infra/k8s/argo/train-dqn-template.yaml
+++ /dev/null
@@ -1,105 +0,0 @@
-# Train DQN: compile → hyperopt → train → evaluate → upload.
-#
-# Usage:
-#   argo submit --from=wftmpl/train-dqn
-#   argo submit --from=wftmpl/train-dqn -p commit-ref=my-branch
-#
-# Thin wrapper around compile-and-train with DQN-tuned defaults.
-# Compiles from source (no pre-built binary dependency).
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: train-dqn
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: train-dqn
-    app.kubernetes.io/part-of: foxhunt
-spec:
-  entrypoint: pipeline
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: compile-and-train
-  securityContext:
-    fsGroup: 1000
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 28800  # 8 hours
-
-  arguments:
-    parameters:
-      - name: model
-        value: dqn
-      - name: commit-sha
-        value: HEAD
-      - name: git-branch
-        value: main
-      - name: gpu-pool
-        value: ci-training-h100
-      - name: hyperopt-trials
-        value: "50"
-      - name: hyperopt-epochs
-        value: "20"
-      - name: train-epochs
-        value: "200"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: feature-cache-dir
-        value: /feature-cache
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: min-hold-bars
-        value: "5"
-      - name: initial-capital
-        value: "35000"
-      - name: ensemble-top-k
-        value: "5"
-      - name: cuda-compute-cap
-        value: "90"
-
-  # Volumes required by compile-and-train steps
-  volumes:
-    - name: git-ssh-key
-      secret:
-        secretName: argo-git-ssh-key
-        defaultMode: 256
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-        readOnly: true
-    - name: cargo-target-cuda
-      persistentVolumeClaim:
-        claimName: cargo-target-cuda
-    - name: feature-cache
-      persistentVolumeClaim:
-        claimName: feature-cache-pvc
-
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: pipeline
-      steps:
-        - - name: compile-and-train
-            templateRef:
-              name: compile-and-train
-              template: pipeline
diff --git a/infra/k8s/argo/train-ppo-template.yaml b/infra/k8s/argo/train-ppo-template.yaml
deleted file mode 100644
index 0b2a5c9b9..000000000
--- a/infra/k8s/argo/train-ppo-template.yaml
+++ /dev/null
@@ -1,107 +0,0 @@
-# Train PPO: fetch binaries → hyperopt → train → evaluate → upload.
-#
-# Usage:
-#   argo submit --from=wftmpl/train-ppo -p binary-tag=dev-abc1234
-#   argo submit --from=wftmpl/train-ppo -p binary-tag=latest -p train-epochs=100
-#
-# Thin wrapper around training-pipeline (templateRef) with PPO-tuned defaults.
-# PPO uses 45 factored actions (5 exposure × 3 order × 3 urgency).
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: train-ppo
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: train-ppo
-    app.kubernetes.io/part-of: foxhunt
-    app.kubernetes.io/component: gpu-test
-spec:
-  entrypoint: pipeline
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: gpu-test
-      app.kubernetes.io/component: training
-  securityContext:
-    runAsUser: 1000
-    runAsGroup: 1000
-    fsGroup: 1000
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 28800  # 8 hours
-
-  arguments:
-    parameters:
-      - name: model
-        value: ppo
-      - name: commit-sha
-        value: HEAD
-      - name: git-branch
-        value: main
-      - name: gpu-pool
-        value: ci-training-h100
-      - name: hyperopt-trials
-        value: "20"
-      - name: hyperopt-epochs
-        value: "8"
-      - name: train-epochs
-        value: "50"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: initial-capital
-        value: "35000"
-      - name: feature-cache-dir
-        value: /feature-cache
-      - name: ensemble-top-k
-        value: "5"
-      - name: cuda-compute-cap
-        value: "90"
-
-  # Volumes required by compile-and-train steps
-  volumes:
-    - name: git-ssh-key
-      secret:
-        secretName: argo-git-ssh-key
-        defaultMode: 256
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-        readOnly: true
-    - name: cargo-target-cuda
-      persistentVolumeClaim:
-        claimName: cargo-target-cuda
-    - name: feature-cache
-      persistentVolumeClaim:
-        claimName: feature-cache-pvc
-
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: pipeline
-      steps:
-        - - name: compile-and-train
-            templateRef:
-              name: compile-and-train
-              template: pipeline
diff --git a/infra/k8s/argo/train-supervised-template.yaml b/infra/k8s/argo/train-supervised-template.yaml
deleted file mode 100644
index 87ca1d557..000000000
--- a/infra/k8s/argo/train-supervised-template.yaml
+++ /dev/null
@@ -1,111 +0,0 @@
-# Train supervised model: fetch binaries → hyperopt → train → evaluate → upload.
-#
-# Usage (any of the 8 supervised models):
-#   argo submit --from=wftmpl/train-supervised -p model=tft -p binary-tag=latest
-#   argo submit --from=wftmpl/train-supervised -p model=mamba2 -p binary-tag=dev-abc1234
-#   argo submit --from=wftmpl/train-supervised -p model=kan -p binary-tag=latest -p train-epochs=100
-#
-# Supported models: tft, mamba2, tggn, tlob, liquid, kan, xlstm, diffusion
-#
-# Thin wrapper around training-pipeline (templateRef) with supervised defaults.
-# Uses hyperopt_baseline_supervised / train_baseline_supervised / evaluate_supervised.
----
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: train-supervised
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: train-supervised
-    app.kubernetes.io/part-of: foxhunt
-    app.kubernetes.io/component: gpu-test
-spec:
-  entrypoint: pipeline
-  serviceAccountName: argo-workflow
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: gpu-test
-      app.kubernetes.io/component: training
-  securityContext:
-    runAsUser: 1000
-    runAsGroup: 1000
-    fsGroup: 1000
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 28800  # 8 hours
-
-  arguments:
-    parameters:
-      - name: model
-        # One of: tft, mamba2, tggn, tlob, liquid, kan, xlstm, diffusion
-        value: tft
-      - name: commit-sha
-        value: HEAD
-      - name: git-branch
-        value: main
-      - name: gpu-pool
-        value: ci-training-h100
-      - name: hyperopt-trials
-        value: "20"
-      - name: hyperopt-epochs
-        value: "8"
-      - name: train-epochs
-        value: "50"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: initial-capital
-        value: "35000"
-      - name: feature-cache-dir
-        value: /feature-cache
-      - name: ensemble-top-k
-        value: "5"
-      - name: cuda-compute-cap
-        value: "90"
-
-  # Volumes required by compile-and-train steps
-  volumes:
-    - name: git-ssh-key
-      secret:
-        secretName: argo-git-ssh-key
-        defaultMode: 256
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-        readOnly: true
-    - name: cargo-target-cuda
-      persistentVolumeClaim:
-        claimName: cargo-target-cuda
-    - name: feature-cache
-      persistentVolumeClaim:
-        claimName: feature-cache-pvc
-
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: pipeline
-      steps:
-        - - name: compile-and-train
-            templateRef:
-              name: compile-and-train
-              template: pipeline
diff --git a/infra/k8s/argo/training-workflow-template.yaml b/infra/k8s/argo/training-workflow-template.yaml
deleted file mode 100644
index 680b9aaee..000000000
--- a/infra/k8s/argo/training-workflow-template.yaml
+++ /dev/null
@@ -1,489 +0,0 @@
-apiVersion: argoproj.io/v1alpha1
-kind: WorkflowTemplate
-metadata:
-  name: training-pipeline
-  namespace: foxhunt
-  labels:
-    app.kubernetes.io/name: training-pipeline
-    app.kubernetes.io/part-of: foxhunt
-    app.kubernetes.io/component: gpu-test
-spec:
-  entrypoint: train-model
-  serviceAccountName: argo-workflow
-  # Label all workflow pods so network policies (MinIO, DNS) allow traffic
-  podMetadata:
-    labels:
-      app.kubernetes.io/part-of: foxhunt
-      app.kubernetes.io/component: training-workflow
-  securityContext:
-    runAsUser: 1000
-    runAsGroup: 1000
-    fsGroup: 1000
-  ttlStrategy:
-    secondsAfterCompletion: 3600
-  activeDeadlineSeconds: 21600
-
-  arguments:
-    parameters:
-      - name: model
-      - name: binary-tag
-        # Required: full commit SHA, e.g. "dev-6ab9e184abcd1234..."
-        # No default — caller must specify which compiled binaries to use.
-      - name: gpu-pool
-        value: ci-training-h100  # Pools: ci-training-h100, ci-training-l40s, ci-training-h100x2
-      - name: hyperopt-trials
-        value: "20"
-      - name: hyperopt-epochs
-        value: "8"
-      - name: train-epochs
-        value: "50"
-      - name: symbol
-        value: ES.FUT
-      - name: data-dir
-        value: /data/futures-baseline
-      - name: mbp10-data-dir
-        value: /data/futures-baseline-mbp10
-      - name: trades-data-dir
-        value: /data/futures-baseline-trades
-      - name: tx-cost-bps
-        value: "0.1"
-      - name: tick-size
-        value: "0.25"
-      - name: spread-ticks
-        value: "1.0"
-      - name: initial-capital
-        value: "35000"
-      - name: ensemble-top-k
-        value: "5"
-      - name: cuda-compute-cap
-        value: "90"
-
-  # Training data PVC — already populated by download jobs (OHLCV, MBP-10, trades).
-  # RWO: only one pod at a time, but steps run sequentially so no conflict.
-  volumes:
-    - name: training-data
-      persistentVolumeClaim:
-        claimName: training-data-pvc
-        readOnly: true
-
-  # Ephemeral shared storage for passing artifacts between DAG steps.
-  # Auto-deleted when workflow completes (all artifacts uploaded to MinIO).
-  volumeClaimTemplates:
-    - metadata:
-        name: workspace
-      spec:
-        accessModes: ["ReadWriteOnce"]
-        storageClassName: scw-bssd
-        resources:
-          requests:
-            storage: 5Gi
-
-  templates:
-    - name: train-model
-      dag:
-        tasks:
-          - name: fetch-binary
-            template: fetch-binary
-          # gpu-warmup disabled — re-enable when GPU nodes stay warm
-          # - name: gpu-warmup
-          #   template: gpu-warmup
-          - name: hyperopt
-            template: hyperopt
-            dependencies: [fetch-binary]
-          - name: train-best
-            template: train-best
-            dependencies: [hyperopt]
-          - name: evaluate
-            template: evaluate
-            dependencies: [train-best]
-          - name: upload-results
-            template: upload-results
-            dependencies: [evaluate]
-
-    # ── gpu-warmup: disabled — re-enable when GPU nodes stay warm ──
-    # - name: gpu-warmup
-    #   nodeSelector:
-    #     k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-    #   tolerations:
-    #     - key: nvidia.com/gpu
-    #       operator: Exists
-    #       effect: NoSchedule
-    #     - key: node.cilium.io/agent-not-ready
-    #       operator: Exists
-    #       effect: NoSchedule
-    #   container:
-    #     image: busybox:1.37
-    #     command: ["/bin/sh", "-c"]
-    #     args:
-    #       - |
-    #         echo "GPU warmup: triggering node autoscale..."
-    #         echo "GPU node scheduled, exiting to free resources"
-    #     resources:
-    #       requests:
-    #         nvidia.com/gpu: "1"
-    #         cpu: 100m
-    #         memory: 64Mi
-    #       limits:
-    #         nvidia.com/gpu: "1"
-    #         cpu: 200m
-    #         memory: 128Mi
-
-    - name: fetch-binary
-      nodeSelector:
-        k8s.scaleway.com/pool-name: platform
-      container:
-        image: curlimages/curl:8.12.1
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            mkdir -p /workspace/bin
-            GITLAB_API="http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181/api/v4"
-
-            TAG="{{workflow.parameters.binary-tag}}"
-            if [ -z "$TAG" ]; then
-              echo "ERROR: binary-tag parameter is required (full commit SHA, e.g. dev-6ab9e184...)"
-              exit 1
-            fi
-            echo "Fetching training binaries from GitLab package: foxhunt-training/${TAG}"
-
-            BINARIES="hyperopt_baseline_rl hyperopt_baseline_supervised train_baseline_rl train_baseline_supervised evaluate_baseline evaluate_supervised training_uploader"
-            for bin in $BINARIES; do
-              echo "  Downloading ${bin}..."
-              curl -fSL -o "/workspace/bin/${bin}" \
-                -H "PRIVATE-TOKEN: ${GITLAB_PAT}" \
-                "${GITLAB_API}/projects/1/packages/generic/foxhunt-training/${TAG}/${bin}" || {
-                echo "WARN: ${bin} not found in package foxhunt-training/${TAG}, skipping"
-                continue
-              }
-            done
-
-            chmod +x /workspace/bin/*
-            echo "Fetched binaries:"
-            ls -lh /workspace/bin/
-        env:
-          - name: GITLAB_PAT
-            valueFrom:
-              secretKeyRef:
-                name: gitlab-pat
-                key: token
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-        resources:
-          requests:
-            cpu: 100m
-            memory: 128Mi
-          limits:
-            cpu: 500m
-            memory: 256Mi
-
-    - name: hyperopt
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-            nvidia-smi
-
-            # Training data available via training-data-pvc mounted at /data
-            echo "=== Training data (PVC) ==="
-            echo "OHLCV: $(ls {{workflow.parameters.data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-            echo "MBP-10: $(ls {{workflow.parameters.mbp10-data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-            echo "Trades: $(ls {{workflow.parameters.trades-data-dir}}/ES.FUT/*.dbn.zst 2>/dev/null | wc -l) files"
-
-            mkdir -p /workspace/output/hyperopt
-
-            MODEL="{{workflow.parameters.model}}"
-            case "$MODEL" in
-              dqn|ppo) BINARY=hyperopt_baseline_rl ;;
-              *)       BINARY=hyperopt_baseline_supervised ;;
-            esac
-
-            TRIALS={{workflow.parameters.hyperopt-trials}}
-            N_INITIAL=5
-            if [ "$TRIALS" -le "$N_INITIAL" ]; then
-              N_INITIAL=$((TRIALS > 1 ? TRIALS - 1 : 1))
-            fi
-
-            echo "Running $BINARY --model $MODEL ($TRIALS trials, $N_INITIAL initial x {{workflow.parameters.hyperopt-epochs}} epochs)"
-            # Phase 1 (fast): fix architecture, search learning dynamics
-            $BINARY \
-              --model "$MODEL" \
-              --phase fast \
-              --trials "$TRIALS" \
-              --n-initial "$N_INITIAL" \
-              --epochs {{workflow.parameters.hyperopt-epochs}} \
-              --parallel 0 \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              --initial-capital {{workflow.parameters.initial-capital}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --mbp10-data-dir {{workflow.parameters.mbp10-data-dir}} \
-              --trades-data-dir {{workflow.parameters.trades-data-dir}} \
-              --base-dir /workspace/output/hyperopt \
-              --output /workspace/output/${MODEL}_phase1_results.json
-
-            echo "=== Phase 1 complete ==="
-            cat /workspace/output/${MODEL}_phase1_results.json 2>/dev/null || echo "No Phase 1 results"
-
-            # Phase 2 (full): fix dynamics from Phase 1, search architecture
-            PHASE2_TRIALS=$((TRIALS / 2))
-            [ "$PHASE2_TRIALS" -lt 5 ] && PHASE2_TRIALS=5
-            PHASE2_EPOCHS=$(({{workflow.parameters.hyperopt-epochs}} * 2))
-
-            $BINARY \
-              --model "$MODEL" \
-              --phase full \
-              --hyperopt-params /workspace/output/${MODEL}_phase1_results.json \
-              --trials $PHASE2_TRIALS \
-              --epochs $PHASE2_EPOCHS \
-              --parallel 0 \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              --initial-capital {{workflow.parameters.initial-capital}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --mbp10-data-dir {{workflow.parameters.mbp10-data-dir}} \
-              --trades-data-dir {{workflow.parameters.trades-data-dir}} \
-              --base-dir /workspace/output/hyperopt \
-              --output /workspace/output/${MODEL}_hyperopt_results.json
-
-            echo "=== Two-phase hyperopt complete ==="
-            cat /workspace/output/${MODEL}_hyperopt_results.json 2>/dev/null || echo "No results file"
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-          - name: OTEL_EXPORTER_OTLP_ENDPOINT
-            value: "http://tempo.foxhunt.svc.cluster.local:4317"
-          - name: CUBLAS_WORKSPACE_CONFIG
-            value: ":4096:8"
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "7"
-            memory: 16Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "8"
-            memory: 48Gi
-
-    - name: train-best
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-            nvidia-smi
-
-            # Training data available via training-data-pvc mounted at /data
-
-            MODEL="{{workflow.parameters.model}}"
-            case "$MODEL" in
-              dqn|ppo) BINARY=train_baseline_rl ;;
-              *)       BINARY=train_baseline_supervised ;;
-            esac
-
-            HYPEROPT_FILE="/workspace/output/${MODEL}_hyperopt_results.json"
-            HYPEROPT_FLAG=""
-            if [ -f "$HYPEROPT_FILE" ]; then
-              HYPEROPT_FLAG="--hyperopt-params $HYPEROPT_FILE"
-            fi
-
-            echo "Training $MODEL with best params ({{workflow.parameters.train-epochs}} epochs)"
-            $BINARY \
-              --model "$MODEL" \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --mbp10-data-dir {{workflow.parameters.mbp10-data-dir}} \
-              --trades-data-dir {{workflow.parameters.trades-data-dir}} \
-              --output-dir /workspace/output \
-              --epochs {{workflow.parameters.train-epochs}} \
-              $HYPEROPT_FLAG \
-              --ensemble-top-k {{workflow.parameters.ensemble-top-k}}
-
-            echo "=== Training complete ==="
-            ls -lh /workspace/output/
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-          - name: FOXHUNT_TRAINING_PROFILE
-            value: /workspace/config/training/dqn-production.toml
-          - name: OTEL_EXPORTER_OTLP_ENDPOINT
-            value: "http://tempo.foxhunt.svc.cluster.local:4317"
-          - name: CUBLAS_WORKSPACE_CONFIG
-            value: ":4096:8"
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "7"
-            memory: 16Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "8"
-            memory: 48Gi
-
-    - name: evaluate
-      nodeSelector:
-        k8s.scaleway.com/pool-name: "{{workflow.parameters.gpu-pool}}"
-      tolerations:
-        - key: nvidia.com/gpu
-          operator: Exists
-          effect: NoSchedule
-        - key: node.cilium.io/agent-not-ready
-          operator: Exists
-          effect: NoSchedule
-      container:
-        image: gitlab-registry.foxhunt.svc.cluster.local:5000/root/foxhunt/ci-builder:latest
-        imagePullPolicy: IfNotPresent
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            export PATH="/workspace/bin:$PATH"
-            export LD_LIBRARY_PATH=$(echo "$LD_LIBRARY_PATH" | tr ':' '\n' | grep -v stubs | tr '\n' ':' | sed 's/:$//')
-
-            # Training data available via training-data-pvc mounted at /data
-
-            MODEL="{{workflow.parameters.model}}"
-            HYPEROPT_FILE="/workspace/output/${MODEL}_hyperopt_results.json"
-            HYPEROPT_FLAG=""
-            if [ -f "$HYPEROPT_FILE" ]; then
-              HYPEROPT_FLAG="--hyperopt-params $HYPEROPT_FILE"
-            fi
-
-            # Evaluate all ensemble members (evaluator auto-discovers *_ensemble_*.safetensors)
-            echo "Evaluating $MODEL"
-            evaluate_baseline \
-              --model "$MODEL" \
-              --models-dir /workspace/output \
-              --data-dir {{workflow.parameters.data-dir}} \
-              --output /workspace/output/${MODEL}_eval_report.json \
-              --symbol {{workflow.parameters.symbol}} \
-              --tx-cost-bps {{workflow.parameters.tx-cost-bps}} \
-              --tick-size {{workflow.parameters.tick-size}} \
-              --spread-ticks {{workflow.parameters.spread-ticks}} \
-              $HYPEROPT_FLAG \
-            || true
-
-            echo "=== Eval report ==="
-            cat /workspace/output/${MODEL}_eval_report.json 2>/dev/null || echo "No eval report"
-        env:
-          - name: RUST_LOG
-            value: info
-          - name: SQLX_OFFLINE
-            value: "true"
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-          - name: training-data
-            mountPath: /data
-            readOnly: true
-        resources:
-          requests:
-            nvidia.com/gpu: "1"
-            cpu: "4"
-            memory: 16Gi
-          limits:
-            nvidia.com/gpu: "1"
-            cpu: "8"
-            memory: 32Gi
-
-    - name: upload-results
-      nodeSelector:
-        k8s.scaleway.com/pool-name: platform
-      container:
-        image: curlimages/curl:8.12.1
-        command: ["/bin/sh", "-c"]
-        args:
-          - |
-            set -e
-            MODEL="{{workflow.parameters.model}}"
-            SYMBOL="{{workflow.parameters.symbol}}"
-            TIMESTAMP=$(date +%Y%m%d-%H%M%S)
-            GITLAB_API="http://gitlab-webservice-default.foxhunt.svc.cluster.local:8181/api/v4"
-            PKG_NAME="foxhunt-training-results"
-            PKG_VERSION="${MODEL}-${SYMBOL}-${TIMESTAMP}"
-
-            echo "Uploading training artifacts to GitLab packages: ${PKG_NAME}/${PKG_VERSION}"
-            echo "  checkpoints (*.safetensors), norm stats, hyperopt results, eval reports"
-
-            UPLOADED=0
-            find /workspace/output -type f | while read -r file; do
-              REL_PATH="${file#/workspace/output/}"
-              # Replace / with -- for flat package file naming
-              SAFE_NAME=$(echo "$REL_PATH" | tr '/' '--')
-              echo "  Uploading ${REL_PATH} as ${SAFE_NAME}..."
-              curl -f --upload-file "$file" \
-                -H "PRIVATE-TOKEN: ${GITLAB_PAT}" \
-                "${GITLAB_API}/projects/1/packages/generic/${PKG_NAME}/${PKG_VERSION}/${SAFE_NAME}" && \
-                UPLOADED=$((UPLOADED + 1)) || \
-                echo "  WARN: Failed to upload ${REL_PATH}"
-            done
-
-            echo "=== Upload complete (${UPLOADED} files) ==="
-            echo "Package: ${PKG_NAME}/${PKG_VERSION}"
-        env:
-          - name: GITLAB_PAT
-            valueFrom:
-              secretKeyRef:
-                name: gitlab-pat
-                key: token
-        volumeMounts:
-          - name: workspace
-            mountPath: /workspace
-        resources:
-          requests:
-            cpu: 100m
-            memory: 128Mi
-          limits:
-            cpu: 500m
-            memory: 256Mi
diff --git a/scripts/argo-precompute.sh b/scripts/argo-precompute.sh
deleted file mode 100755
index 852a1276a..000000000
--- a/scripts/argo-precompute.sh
+++ /dev/null
@@ -1,70 +0,0 @@
-#!/usr/bin/env bash
-# Pre-compute DQN features via Argo Workflows.
-#
-# Usage:
-#   ./scripts/argo-precompute.sh ES.FUT
-#   ./scripts/argo-precompute.sh ES.FUT --pool platform --watch
-#
-# Requires: argo CLI
-set -euo pipefail
-
-TEMPLATE="precompute-features"
-POOL=""
-DATA_DIR=""
-MBP10_DIR=""
-TRADES_DIR=""
-OUTPUT_DIR=""
-WATCH=false
-
-usage() {
-  cat <<EOF
-Usage: $(basename "$0") <symbol> [OPTIONS]
-
-Options:
-  --pool <name>        Node pool (default: ci-compile-cpu)
-  --data-dir <dir>     OHLCV data dir (default: /data/futures-baseline)
-  --mbp10-dir <dir>    MBP-10 data dir (default: /data/futures-baseline-mbp10)
-  --trades-dir <dir>   Trades data dir (default: /data/futures-baseline-trades)
-  --output-dir <dir>   Output dir (default: /data/feature-cache)
-  --watch              Follow workflow logs after submission
-  -h, --help           Show this help
-EOF
-  exit 0
-}
-
-[[ $# -eq 0 ]] && { echo "Error: symbol argument required"; usage; }
-[[ "$1" == "-h" || "$1" == "--help" ]] && usage
-
-SYMBOL="$1"; shift
-
-while [[ $# -gt 0 ]]; do
-  case $1 in
-    --pool)       POOL="$2"; shift 2 ;;
-    --data-dir)   DATA_DIR="$2"; shift 2 ;;
-    --mbp10-dir)  MBP10_DIR="$2"; shift 2 ;;
-    --trades-dir) TRADES_DIR="$2"; shift 2 ;;
-    --output-dir) OUTPUT_DIR="$2"; shift 2 ;;
-    --watch)      WATCH=true; shift ;;
-    -h|--help)    usage ;;
-    *)            echo "Unknown option: $1"; usage ;;
-  esac
-done
-
-CMD="argo submit -n foxhunt --from=wftmpl/$TEMPLATE"
-CMD="$CMD -p symbol=$SYMBOL"
-
-[[ -n "$POOL" ]]       && CMD="$CMD -p node-pool=$POOL"
-[[ -n "$DATA_DIR" ]]   && CMD="$CMD -p data-dir=$DATA_DIR"
-[[ -n "$MBP10_DIR" ]]  && CMD="$CMD -p mbp10-dir=$MBP10_DIR"
-[[ -n "$TRADES_DIR" ]] && CMD="$CMD -p trades-dir=$TRADES_DIR"
-[[ -n "$OUTPUT_DIR" ]] && CMD="$CMD -p output-dir=$OUTPUT_DIR"
-
-if $WATCH; then
-  CMD="$CMD --watch"
-fi
-
-echo "Submitting feature pre-compute workflow..."
-echo "  symbol: $SYMBOL"
-[[ -n "$POOL" ]] && echo "  pool:   $POOL"
-echo ""
-eval "$CMD"
diff --git a/scripts/argo-train.sh b/scripts/argo-train.sh
index ba6aae075..22b5668dc 100755
--- a/scripts/argo-train.sh
+++ b/scripts/argo-train.sh
@@ -1,29 +1,30 @@
 #!/usr/bin/env bash
-# Train a model via Argo Workflows (DQN, PPO, or any supervised model).
+# Train a model via Argo Workflows.
 #
 # Usage:
-#   ./scripts/argo-train.sh dqn                           # DQN with defaults
-#   ./scripts/argo-train.sh ppo --tag dev-abc1234          # PPO with specific binary
-#   ./scripts/argo-train.sh tft --trials 40 --epochs 100   # supervised TFT
-#   ./scripts/argo-train.sh mamba2 --webhook               # trigger via webhook
+#   ./scripts/argo-train.sh dqn                          # defaults: HEAD, H100, 50 epochs
+#   ./scripts/argo-train.sh dqn --sha abc1234            # specific commit
+#   ./scripts/argo-train.sh dqn --epochs 100 --trials 40 # override training params
+#   ./scripts/argo-train.sh ppo --gpu-pool ci-training    # L40S instead of H100
+#   ./scripts/argo-train.sh dqn --baseline                # skip hyperopt
+#   ./scripts/argo-train.sh dqn --watch                  # follow logs
 #
 # Supported models:
 #   RL:          dqn, ppo
 #   Supervised:  tft, mamba2, tggn, tlob, liquid, kan, xlstm, diffusion
 #
-# Requires: argo CLI (default) or curl (webhook mode)
+# Requires: argo CLI
 set -euo pipefail
 
-TAG="latest"
+SHA="HEAD"
+BRANCH="main"
 TRIALS=""
 EPOCHS=""
 GPU_POOL=""
 SYMBOL=""
-WEBHOOK=false
-WEBHOOK_BASE="http://workflow-trigger-eventsource-svc.foxhunt:12001"
 WATCH=false
 BASELINE=false
-CACHE_DIR=""
+CAPITAL=""
 
 usage() {
   cat <<EOF
@@ -31,19 +32,19 @@ Usage: $(basename "$0") <model> [OPTIONS]
 
 Models:
   dqn, ppo                                          (RL)
-  tft, mamba2, tggn, tlob, liquid, kan, xlstm, diffusion  (supervised)
+  tft, mamba2, tggn, tlob, liquid, kan, xlstm       (supervised)
 
 Options:
-  --tag <binary-tag>     Binary tag to use (default: latest)
-  --trials <n>           Hyperopt trials (default: model-specific)
-  --epochs <n>           Training epochs (default: model-specific)
-  --gpu-pool <pool>      GPU node pool (default: ci-training-h100)
-  --symbol <sym>         Trading symbol (default: ES.FUT)
-  --baseline             Skip hyperopt (set trials=0), train with defaults
-  --cache-dir <path>     Override feature cache directory
-  --webhook              Trigger via webhook instead of argo CLI
-  --watch                Follow workflow logs after submission
-  -h, --help             Show this help
+  --sha <commit>       Git commit SHA (default: HEAD)
+  --branch <branch>    Git branch (default: main)
+  --trials <n>         Hyperopt trials (default: 20)
+  --epochs <n>         Training epochs (default: 50)
+  --gpu-pool <pool>    GPU node pool (default: ci-training-h100)
+  --symbol <sym>       Trading symbol (default: ES.FUT)
+  --capital <n>        Initial capital (default: 35000)
+  --baseline           Skip hyperopt (trials=0)
+  --watch              Follow workflow logs
+  -h, --help           Show this help
 EOF
   exit 0
 }
@@ -52,7 +53,6 @@ EOF
 
 MODEL="$1"; shift
 
-# Validate model name
 case "$MODEL" in
   dqn|ppo|tft|mamba2|tggn|tlob|liquid|kan|xlstm|diffusion) ;;
   *) echo "Error: unknown model '$MODEL'"; usage ;;
@@ -60,72 +60,41 @@ esac
 
 while [[ $# -gt 0 ]]; do
   case $1 in
-    --tag)       TAG="$2"; shift 2 ;;
+    --sha)       SHA="$2"; shift 2 ;;
+    --branch)    BRANCH="$2"; shift 2 ;;
     --trials)    TRIALS="$2"; shift 2 ;;
     --epochs)    EPOCHS="$2"; shift 2 ;;
     --gpu-pool)  GPU_POOL="$2"; shift 2 ;;
     --symbol)    SYMBOL="$2"; shift 2 ;;
+    --capital)   CAPITAL="$2"; shift 2 ;;
     --baseline)  BASELINE=true; shift ;;
-    --cache-dir) CACHE_DIR="$2"; shift 2 ;;
-    --webhook)   WEBHOOK=true; shift ;;
     --watch)     WATCH=true; shift ;;
     -h|--help)   usage ;;
     *)           echo "Unknown option: $1"; usage ;;
   esac
 done
 
-# Determine workflow template name
-case "$MODEL" in
-  dqn)  TEMPLATE="train-dqn" ;;
-  ppo)  TEMPLATE="train-ppo" ;;
-  *)    TEMPLATE="train-supervised" ;;
-esac
-
-if $WEBHOOK; then
-  case "$MODEL" in
-    dqn) ENDPOINT="/train/dqn";        PAYLOAD="{\"binary_tag\": \"${TAG}\"}" ;;
-    ppo) ENDPOINT="/train/ppo";        PAYLOAD="{\"binary_tag\": \"${TAG}\"}" ;;
-    *)   ENDPOINT="/train/supervised"; PAYLOAD="{\"model\": \"${MODEL}\", \"binary_tag\": \"${TAG}\"}" ;;
-  esac
-
-  echo "Triggering $MODEL training via webhook..."
-  curl -sf -X POST "${WEBHOOK_BASE}${ENDPOINT}" \
-    -H 'Content-Type: application/json' \
-    -d "$PAYLOAD"
-  echo "Webhook sent: $PAYLOAD"
-  exit 0
-fi
-
-# Build argo submit command
-CMD="argo submit -n foxhunt --from=wftmpl/$TEMPLATE"
-CMD="$CMD -p binary-tag=$TAG"
-
-# Supervised models need --model param
-if [[ "$MODEL" != "dqn" && "$MODEL" != "ppo" ]]; then
-  CMD="$CMD -p model=$MODEL"
-fi
+CMD="argo submit -n foxhunt --from=wftmpl/train"
+CMD="$CMD -p commit-sha=$SHA"
+CMD="$CMD -p git-branch=$BRANCH"
+CMD="$CMD -p model=$MODEL"
 
 [[ -n "$TRIALS" ]]    && CMD="$CMD -p hyperopt-trials=$TRIALS"
 [[ -n "$EPOCHS" ]]    && CMD="$CMD -p train-epochs=$EPOCHS"
 [[ -n "$GPU_POOL" ]]  && CMD="$CMD -p gpu-pool=$GPU_POOL"
 [[ -n "$SYMBOL" ]]    && CMD="$CMD -p symbol=$SYMBOL"
-[[ -n "$CACHE_DIR" ]] && CMD="$CMD -p feature-cache-dir=$CACHE_DIR"
+[[ -n "$CAPITAL" ]]   && CMD="$CMD -p initial-capital=$CAPITAL"
 
-if $BASELINE; then
-  CMD="$CMD -p hyperopt-trials=0"
-fi
+$BASELINE && CMD="$CMD -p hyperopt-trials=0"
+$WATCH    && CMD="$CMD --watch"
 
-if $WATCH; then
-  CMD="$CMD --watch"
-fi
-
-echo "Submitting $MODEL training workflow ($TEMPLATE)..."
-echo "  binary-tag: $TAG"
-$BASELINE                && echo "  mode:       baseline (hyperopt skipped)"
-[[ -n "$TRIALS" ]]       && echo "  trials:     $TRIALS"
-[[ -n "$EPOCHS" ]]       && echo "  epochs:     $EPOCHS"
-[[ -n "$GPU_POOL" ]]     && echo "  gpu-pool:   $GPU_POOL"
-[[ -n "$SYMBOL" ]]       && echo "  symbol:     $SYMBOL"
-[[ -n "$CACHE_DIR" ]]    && echo "  cache-dir:  $CACHE_DIR"
+echo "Submitting $MODEL training workflow..."
+echo "  sha:    $SHA"
+echo "  branch: $BRANCH"
+echo "  model:  $MODEL"
+$BASELINE              && echo "  mode:   baseline (no hyperopt)"
+[[ -n "$TRIALS" ]]     && echo "  trials: $TRIALS"
+[[ -n "$EPOCHS" ]]     && echo "  epochs: $EPOCHS"
+[[ -n "$GPU_POOL" ]]   && echo "  gpu:    $GPU_POOL"
 echo ""
 eval "$CMD"