diff --git a/scripts/alpha-rl-run.sh b/scripts/alpha-rl-run.sh
new file mode 100755
index 000000000..76b47919a
--- /dev/null
+++ b/scripts/alpha-rl-run.sh
@@ -0,0 +1,266 @@
+#!/usr/bin/env bash
+# alpha-rl-run.sh — One-command production training run.
+#
+# Convenience wrapper around argo-alpha-rl.sh that handles the full
+# production flow: push, apply template, submit, monitor, download
+# artifacts, and print a throughput summary.
+#
+# Usage:
+#   ./scripts/alpha-rl-run.sh                    # defaults: 500k steps, b=256, L40S
+#   ./scripts/alpha-rl-run.sh --steps 1000000    # 1M steps
+#   ./scripts/alpha-rl-run.sh --batch 256        # b=256
+#   ./scripts/alpha-rl-run.sh --gpu h100         # H100 pool
+#   ./scripts/alpha-rl-run.sh --nsys             # with nsys profiling
+#   ./scripts/alpha-rl-run.sh --dry-run          # print command, don't submit
+#
+# Flow:
+#   1. Push current branch
+#   2. Apply Argo template
+#   3. Submit workflow
+#   4. Monitor until completion
+#   5. Download JSONL + summary from PVC
+#   6. Print throughput + key metrics
+
+set -euo pipefail
+
+# ── Defaults ────────────────────────────────────────────────────────────
+# Per feedback_default_to_l40s_pool: L40S is the default.
+STEPS=500000
+BATCH=256
+GPU_POOL=ci-training-l40s
+NSYS=false
+BRANCH=$(git branch --show-current 2>/dev/null || echo "ml-alpha-phase-a")
+DRY_RUN=false
+SEED=16962
+SEQ_LEN=32
+PER_CAPACITY=32768
+INSTRUMENT_MODE=all
+FOLD_IDX=0
+N_FOLDS=1
+N_EVAL_STEPS=0
+
+usage() {
+  cat <<EOF
+Usage: $(basename "$0") [OPTIONS]
+
+Production training run — push, submit, monitor, download, summarize.
+
+Options:
+  --steps <n>       Training steps (default: $STEPS)
+  --batch <n>       Parallel LobSim instances per step (default: $BATCH)
+  --gpu <pool>      l40s | h100 (default: l40s)
+  --nsys            Enable nsys profiling
+  --seed <n>        Random seed (default: $SEED)
+  --seq-len <n>     Encoder sequence length (default: $SEQ_LEN)
+  --per-capacity <n> PER buffer capacity (default: $PER_CAPACITY)
+  --instrument <m>  all | front-month | id=<N> (default: $INSTRUMENT_MODE)
+  --fold-idx <k>    Walk-forward fold index (default: $FOLD_IDX)
+  --n-folds <K>     Total fold count (default: $N_FOLDS)
+  --n-eval-steps <n> Eval-phase step count (default: $N_EVAL_STEPS)
+  --branch <b>      Git branch (default: current branch)
+  --dry-run         Print the submit command without executing
+  -h, --help        Show this help
+EOF
+}
+
+while [[ $# -gt 0 ]]; do
+  case "$1" in
+    --steps)         STEPS="$2"; shift 2 ;;
+    --batch)         BATCH="$2"; shift 2 ;;
+    --gpu)
+      case "$2" in
+        l40s)  GPU_POOL=ci-training-l40s ;;
+        h100)  GPU_POOL=ci-training-h100 ;;
+        *)     echo "ERROR: unknown gpu: $2 (use l40s or h100)"; exit 1 ;;
+      esac
+      shift 2 ;;
+    --nsys)          NSYS=true; shift ;;
+    --seed)          SEED="$2"; shift 2 ;;
+    --seq-len)       SEQ_LEN="$2"; shift 2 ;;
+    --per-capacity)  PER_CAPACITY="$2"; shift 2 ;;
+    --instrument)    INSTRUMENT_MODE="$2"; shift 2 ;;
+    --fold-idx)      FOLD_IDX="$2"; shift 2 ;;
+    --n-folds)       N_FOLDS="$2"; shift 2 ;;
+    --n-eval-steps)  N_EVAL_STEPS="$2"; shift 2 ;;
+    --branch)        BRANCH="$2"; shift 2 ;;
+    --dry-run)       DRY_RUN=true; shift ;;
+    -h|--help)       usage; exit 0 ;;
+    *) echo "Unknown option: $1"; usage; exit 1 ;;
+  esac
+done
+
+# ── Step 1: Push current branch ────────────────────────────────────────
+# Per feedback_push_before_deploy: always push before submitting.
+echo "=== Step 1: Push $BRANCH ==="
+if [[ "$DRY_RUN" == "false" ]]; then
+  git push origin "$BRANCH"
+  echo "  pushed: $BRANCH"
+else
+  echo "  [dry-run] would push: git push origin $BRANCH"
+fi
+
+SHA=$(git rev-parse --short=9 HEAD)
+echo "  SHA: $SHA"
+
+# ── Step 2: Apply Argo template ────────────────────────────────────────
+# Per feedback_argo_template_must_apply: always apply before submit.
+TEMPLATE_FILE="infra/k8s/argo/alpha-rl-template.yaml"
+if [[ ! -f "$TEMPLATE_FILE" ]]; then
+  echo "ERROR: template not found at $TEMPLATE_FILE (run from repo root)"
+  exit 1
+fi
+
+echo ""
+echo "=== Step 2: Apply Argo template ==="
+if [[ "$DRY_RUN" == "false" ]]; then
+  kubectl apply -n foxhunt -f "$TEMPLATE_FILE"
+  echo "  applied: $TEMPLATE_FILE"
+else
+  echo "  [dry-run] would apply: kubectl apply -n foxhunt -f $TEMPLATE_FILE"
+fi
+
+# ── GPU pool → SM arch ──────────────────────────────────────────────────
+case "$GPU_POOL" in
+  ci-training-l40s)  SM=89 ;;
+  ci-training-h100)  SM=90 ;;
+esac
+
+# ── Step 3: Submit workflow ─────────────────────────────────────────────
+echo ""
+echo "=== Step 3: Submit workflow ==="
+echo "  branch:       $BRANCH"
+echo "  sha:          $SHA"
+echo "  gpu-pool:     $GPU_POOL (sm_$SM)"
+echo "  steps:        $STEPS"
+echo "  batch:        $BATCH"
+echo "  seq-len:      $SEQ_LEN"
+echo "  per-capacity: $PER_CAPACITY"
+echo "  seed:         $SEED"
+echo "  instrument:   $INSTRUMENT_MODE"
+echo "  nsys:         $NSYS"
+
+SUBMIT_CMD="argo submit -n foxhunt --from=wftmpl/alpha-rl \
+  -p commit-sha=$SHA \
+  -p git-branch=$BRANCH \
+  -p cuda-compute-cap=$SM \
+  -p gpu-pool=$GPU_POOL \
+  -p n-steps=$STEPS \
+  -p seq-len=$SEQ_LEN \
+  -p n-backtests=$BATCH \
+  -p per-capacity=$PER_CAPACITY \
+  -p seed=$SEED \
+  -p instrument-mode=$INSTRUMENT_MODE \
+  -p fold-idx=$FOLD_IDX \
+  -p n-folds=$N_FOLDS \
+  -p n-eval-steps=$N_EVAL_STEPS \
+  -p nsys-profile=$NSYS"
+
+if [[ "$DRY_RUN" == "true" ]]; then
+  echo ""
+  echo "[dry-run] Would submit:"
+  echo "  $SUBMIT_CMD"
+  exit 0
+fi
+
+WF_OUTPUT=$(eval "$SUBMIT_CMD" 2>&1)
+echo "$WF_OUTPUT"
+
+# Extract workflow name from argo submit output
+WF_NAME=$(echo "$WF_OUTPUT" | grep -oP 'Name:\s+\K\S+' || echo "")
+if [[ -z "$WF_NAME" ]]; then
+  echo "ERROR: could not extract workflow name from argo submit output"
+  echo "       Check the output above and monitor manually."
+  exit 1
+fi
+echo ""
+echo "  workflow: $WF_NAME"
+
+# ── Step 4: Monitor until completion ────────────────────────────────────
+echo ""
+echo "=== Step 4: Monitoring $WF_NAME ==="
+echo "  (polling every 30s — Ctrl-C to detach, workflow continues)"
+echo ""
+
+START_TS=$(date +%s)
+
+while true; do
+  WF_PHASE=$(argo get -n foxhunt "$WF_NAME" -o json 2>/dev/null \
+    | grep -oP '"phase"\s*:\s*"\K[^"]+' | head -1 || echo "Pending")
+
+  case "$WF_PHASE" in
+    Succeeded)
+      echo ""
+      echo "  SUCCEEDED after $(( ($(date +%s) - START_TS) / 60 ))m"
+      break
+      ;;
+    Failed|Error)
+      echo ""
+      echo "  FAILED ($WF_PHASE) after $(( ($(date +%s) - START_TS) / 60 ))m"
+      echo ""
+      echo "=== Failure logs ==="
+      argo logs -n foxhunt "$WF_NAME" --tail 50 2>/dev/null || true
+      exit 1
+      ;;
+    *)
+      printf "\r  status: %-12s elapsed: %dm" "$WF_PHASE" "$(( ($(date +%s) - START_TS) / 60 ))"
+      sleep 30
+      ;;
+  esac
+done
+
+# ── Step 5: Download JSONL + summary from PVC ───────────────────────────
+echo ""
+echo "=== Step 5: Download artifacts ==="
+
+# Find the pod that ran the workflow
+POD=$(kubectl get pods -n foxhunt -l "workflows.argoproj.io/workflow=$WF_NAME" \
+  --sort-by=.metadata.creationTimestamp -o jsonpath='{.items[-1].metadata.name}' 2>/dev/null || echo "")
+
+OUT_DIR="/feature-cache/alpha-rl-runs/$SHA/fold$FOLD_IDX"
+LOCAL_DIR="runs/$SHA"
+mkdir -p "$LOCAL_DIR"
+
+# Print final logs
+echo ""
+echo "=== Step 5a: Final training logs ==="
+argo logs -n foxhunt "$WF_NAME" --tail 30 2>/dev/null \
+  | grep -E "^step|complete|SUMMARY|throughput|sharpe|pnl" || true
+
+# Copy summary JSON if the pod is still around
+if [[ -n "$POD" ]]; then
+  kubectl cp "foxhunt/$POD:$OUT_DIR/alpha_rl_train_summary.json" \
+    "$LOCAL_DIR/alpha_rl_train_summary.json" 2>/dev/null || true
+  kubectl cp "foxhunt/$POD:$OUT_DIR/diagnostics.jsonl" \
+    "$LOCAL_DIR/diagnostics.jsonl" 2>/dev/null || true
+  if [[ "$NSYS" == "true" ]]; then
+    kubectl cp "foxhunt/$POD:$OUT_DIR/nsys_trace.nsys-rep" \
+      "$LOCAL_DIR/nsys_trace.nsys-rep" 2>/dev/null || true
+    kubectl cp "foxhunt/$POD:$OUT_DIR/nsys_trace.sqlite" \
+      "$LOCAL_DIR/nsys_trace.sqlite" 2>/dev/null || true
+  fi
+  echo "  downloaded to: $LOCAL_DIR/"
+  ls -lh "$LOCAL_DIR/" 2>/dev/null || true
+else
+  echo "  WARNING: pod not found — artifacts remain on PVC at $OUT_DIR"
+  echo "           Use: kubectl cp foxhunt/<pod>:$OUT_DIR/ $LOCAL_DIR/"
+fi
+
+# ── Step 6: Print throughput + key metrics ──────────────────────────────
+echo ""
+echo "=== Step 6: Summary ==="
+ELAPSED_MIN=$(( ($(date +%s) - START_TS) / 60 ))
+echo "  wall time:  ${ELAPSED_MIN}m"
+echo "  steps:      $STEPS"
+echo "  batch:      $BATCH"
+echo "  gpu:        $GPU_POOL"
+echo "  sha:        $SHA"
+
+if [[ -f "$LOCAL_DIR/alpha_rl_train_summary.json" ]]; then
+  echo ""
+  echo "  --- Training Summary ---"
+  cat "$LOCAL_DIR/alpha_rl_train_summary.json"
+  echo ""
+fi
+
+echo ""
+echo "Done. Artifacts in: $LOCAL_DIR/"