diff --git a/prod/clean_arch/violet/alpha_wrappers.py b/prod/clean_arch/violet/alpha_wrappers.py
new file mode 100644
index 0000000..09fe535
--- /dev/null
+++ b/prod/clean_arch/violet/alpha_wrappers.py
@@ -0,0 +1,214 @@
+"""VIOLET V3a: V-TYPES-bounded wrappers over BLUE's LIVE alpha kernels.
+
+L1 (PURE ALPHA) of the V3 three-layer architecture: decide + size *literally as
+BLUE* by WRAPPING — never reimplementing — the same production kernels that
+``prod/nautilus_event_trader.py`` (``DolphinLiveTrader``) runs:
+
+  - ``AlphaAssetSelector`` (alpha_asset_selector.py) — IRP selection, BIBLE §5
+  - ``AlphaBetSizer``      (alpha_bet_sizer.py)      — cubic-convex conviction
+                                                       leverage + alpha fraction §6
+  - ``AlphaExitEngineV7``  (alpha_exit_v7_engine.py) — live multi-leg exit (as-is)
+
+``prod/clean_arch/dita_v2/blue_parity.py`` (PinkAssetPicker/PinkAlphaSizer) is the
+REFERENCE for the wrap discipline (observe/dedupe-on-scan), NOT authoritative:
+it caps ``max_leverage`` at 8.0 while the live kernel *default* is 5.0 and the
+recorded conviction distribution reaches 9.0. Therefore ``max_leverage`` (and the
+other sizer knobs) are REQUIRED explicit parameters here — the V3d parity harness
+pins them from the recorded data; nothing in VIOLET hardcodes a value the way the
+drifted wrapper did.
+
+DUAL-LEVERAGE: the conviction leverage produced here SIZES THE QUANTITY (the
+internal/conviction side). The exchange-leverage mapping (``prod/bingx/leverage.py``,
+max-3x cubic) is L3 (tradeability), applied downstream — NEVER in this module.
+This module is exchange-agnostic by construction.
+"""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+from typing import Annotated, Any, Dict, List, Literal, Optional, Tuple
+
+from pydantic import Field
+
+from .domain import StrictModel, Symbol, typed
+
+_PROJECT_ROOT = Path(__file__).resolve().parents[3]
+
+# ── refined alpha scalars (L1, exchange-agnostic) ─────────────────────────────
+# ConvictionLeverage is the bet-sizer's internal leverage — it sizes the
+# quantity. It is NOT exchange leverage. Range mirrors the kernel's own
+# [min_leverage, max_leverage] envelope; we keep a generous finite ceiling and
+# let the kernel's clamp be authoritative (the parity harness verifies it).
+ConvictionLeverage = Annotated[float, Field(ge=0.0, le=64.0, allow_inf_nan=False)]
+Fraction = Annotated[float, Field(ge=0.0, le=1.0, allow_inf_nan=False)]
+VelDiv = Annotated[float, Field(allow_inf_nan=False)]
+Side = Literal["LONG", "SHORT"]
+
+
+def _import_blue_alpha() -> Tuple[Any, Any, Any]:
+    """Import BLUE's live selector/sizer/exit-v7 kernels.
+
+    Mirrors blue_parity._import_blue_modules: the real package lives at
+    <root>/nautilus_dolphin/nautilus_dolphin/; BLUE runs with
+    <root>/nautilus_dolphin on sys.path.
+    """
+    try:
+        from nautilus_dolphin.nautilus import (  # type: ignore
+            alpha_asset_selector, alpha_bet_sizer, alpha_exit_v7_engine,
+        )
+    except ImportError:
+        for p in (str(_PROJECT_ROOT / "nautilus_dolphin"), str(_PROJECT_ROOT)):
+            if p not in sys.path:
+                sys.path.insert(0, p)
+        sys.modules.pop("nautilus_dolphin", None)
+        from nautilus_dolphin.nautilus import (  # type: ignore
+            alpha_asset_selector, alpha_bet_sizer, alpha_exit_v7_engine,
+        )
+    return alpha_asset_selector, alpha_bet_sizer, alpha_exit_v7_engine
+
+
+# ── L1 value objects (the pure-alpha output, exchange-agnostic) ────────────────
+
+class AssetPick(StrictModel):
+    """One IRP selection result — the asset the alpha would trade this scan."""
+
+    asset: Symbol
+    side: Side
+    ars_score: float
+    alignment: float
+
+
+class SizeDecision(StrictModel):
+    """Bet-sizer output. ``notional_fraction = fraction * conviction_leverage``
+    is the realized notional/capital (== the recorded ``our_leverage``); it is
+    the conviction side of the dual-leverage and is exchange-agnostic."""
+
+    fraction: Fraction
+    conviction_leverage: ConvictionLeverage
+    notional_fraction: float = Field(ge=0.0, allow_inf_nan=False)
+    bucket_idx: int = Field(ge=0, le=3)
+    strength_score: float = Field(ge=0.0, allow_inf_nan=False)
+    signal_bucket: str
+
+
+# ── wrappers ───────────────────────────────────────────────────────────────────
+
+class VioletAssetSelector:
+    """Wraps BLUE's ``AlphaAssetSelector`` — IRP ranking over the scan universe.
+
+    Faithful to blue_parity.PinkAssetPicker semantics: per-asset price history
+    accumulated from scan payloads, ranked at signal time; no-fallback (§5.5).
+    """
+
+    def __init__(self, lookback_horizon: int = 0, min_alignment: float = 0.0):
+        sel_mod, _, _ = _import_blue_alpha()
+        self._mod = sel_mod
+        self.lookback = int(lookback_horizon) if lookback_horizon > 0 else int(sel_mod.IRP_LOOKBACK)
+        self.min_alignment = float(min_alignment)
+        self._selector = sel_mod.AlphaAssetSelector(lookback_horizon=self.lookback)
+
+    @typed
+    def pick(self, market_data: dict[str, list[float]], regime_direction: int) -> Optional[AssetPick]:
+        """Rank the warm universe; return the first survivor as a typed pick.
+
+        ``market_data``: per-asset price history (>= lookback+1 prices).
+        ``regime_direction``: -1 short regime, +1 long.
+        """
+        if not market_data:
+            return None
+        rankings = self._selector.rank_assets(market_data, regime_direction=int(regime_direction))
+        expected = "SHORT" if int(regime_direction) == -1 else "LONG"
+        for r in rankings:
+            if self.min_alignment > 0 and r.metrics.alignment < self.min_alignment:
+                continue
+            if r.action != expected:
+                continue
+            return AssetPick(
+                asset=str(r.asset).upper(),
+                side=r.action,
+                ars_score=float(r.ars_score),
+                alignment=float(r.metrics.alignment),
+            )
+        return None
+
+
+class VioletBetSizer:
+    """Wraps BLUE's ``AlphaBetSizer`` — cubic-convex conviction leverage + fraction.
+
+    All sizer knobs are EXPLICIT (no hidden defaults): the live kernel default
+    ``max_leverage`` is 5.0, blue_parity passes 8.0, and recorded conviction
+    reaches 9.0 — so the caller (and the parity harness) must pass the value
+    pinned from live BLUE, never inherit a drifted default.
+    """
+
+    def __init__(
+        self,
+        *,
+        base_fraction: float,
+        min_leverage: float,
+        max_leverage: float,
+        leverage_convexity: float = 3.0,
+        vel_div_threshold: float = -0.02,
+        vel_div_extreme: float = -0.05,
+        use_dynamic_leverage: bool = True,
+        use_alpha_layers: bool = True,
+    ):
+        _, sizer_mod, _ = _import_blue_alpha()
+        self._sizer = sizer_mod.AlphaBetSizer(
+            base_fraction=base_fraction,
+            min_leverage=min_leverage,
+            max_leverage=max_leverage,
+            leverage_convexity=leverage_convexity,
+            vel_div_threshold=vel_div_threshold,
+            vel_div_extreme=vel_div_extreme,
+            use_dynamic_leverage=use_dynamic_leverage,
+            use_alpha_layers=use_alpha_layers,
+        )
+        self.max_leverage = float(max_leverage)
+
+    @typed
+    def calculate(
+        self, *, capital: float, vel_div: float,
+        vel_div_trend: float = 0.0, trade_direction: int = -1,
+    ) -> SizeDecision:
+        raw = self._sizer.calculate_size(
+            capital=float(capital), vel_div=float(vel_div),
+            vel_div_trend=float(vel_div_trend), trade_direction=int(trade_direction),
+        )
+        fraction = float(raw["fraction"])
+        leverage = float(raw["leverage"])
+        breakdown = raw.get("breakdown", {})
+        return SizeDecision(
+            fraction=fraction,
+            conviction_leverage=leverage,
+            notional_fraction=fraction * leverage,            # == our_leverage
+            bucket_idx=int(raw.get("bucket_idx", 3)),
+            strength_score=float(breakdown.get("strength_score", 0.0)),
+            signal_bucket=str(breakdown.get("signal_bucket", "")),
+        )
+
+    def record_trade(self, bucket_idx: int, pnl: float) -> None:
+        """Feed realized PnL back into the alpha layers (bucket boost/streak)."""
+        self._sizer.record_trade(int(bucket_idx), float(pnl))
+
+
+class VioletExitEngine:
+    """Wraps BLUE's live ``AlphaExitEngineV7`` (multi-leg exit). Run AS-IS first
+    (operator mandate); refactor/promote/demote later. Exchange-agnostic: emits
+    an exit decision dict from the kernel, not a venue order."""
+
+    def __init__(self, *, bar_duration_sec: float = 11.0, **engine_kwargs: Any):
+        _, _, exit_mod = _import_blue_alpha()
+        self._mod = exit_mod
+        self._engine = exit_mod.AlphaExitEngineV7(
+            bar_duration_sec=bar_duration_sec, **engine_kwargs,
+        )
+
+    def make_context(self, *args: Any, **kwargs: Any) -> Any:
+        return self._engine.make_context(*args, **kwargs)
+
+    @typed
+    def evaluate(self, context: Any) -> dict:
+        """Return the kernel's exit decision dict verbatim (parity-faithful)."""
+        return self._engine.evaluate(context)
diff --git a/prod/clean_arch/violet/test_violet_alpha_wrappers.py b/prod/clean_arch/violet/test_violet_alpha_wrappers.py
new file mode 100644
index 0000000..699cccd
--- /dev/null
+++ b/prod/clean_arch/violet/test_violet_alpha_wrappers.py
@@ -0,0 +1,87 @@
+"""V3a: V-TYPES wrappers over BLUE's live alpha kernels — parity + drift guards."""
+
+from __future__ import annotations
+
+import math
+import sys
+
+sys.path.insert(0, "/mnt/dolphinng5_predict")
+
+import pytest
+from hypothesis import given, settings, strategies as st
+from pydantic import ValidationError
+
+from prod.clean_arch.violet.alpha_wrappers import (
+    AssetPick, SizeDecision, VioletAssetSelector, VioletBetSizer,
+)
+
+
+def _sizer(max_leverage: float = 9.0) -> VioletBetSizer:
+    return VioletBetSizer(base_fraction=0.20, min_leverage=0.5, max_leverage=max_leverage)
+
+
+def _universe(n: int = 4, bars: int = 60) -> dict:
+    return {
+        f"A{i}USDT": [100.0 * (1 + 0.001 * math.sin(0.1 * t) + 0.0005 * i * t) for t in range(bars)]
+        for i in range(n)
+    }
+
+
+# ── sizing reproduces the recorded model ──────────────────────────────────────
+
+def test_notional_fraction_is_fraction_times_conviction():
+    d = _sizer().calculate(capital=69_000.0, vel_div=-0.05, trade_direction=-1)
+    assert d.notional_fraction == pytest.approx(d.fraction * d.conviction_leverage)
+
+
+def test_extreme_short_signal_saturates_at_max_leverage_9():
+    # vel_div at/under extreme drives conviction to the cap; with max_leverage=9
+    # and base_fraction 0.20, notional_fraction == 1.8 — the recorded our_leverage max.
+    d = _sizer(9.0).calculate(capital=69_000.0, vel_div=-0.20, trade_direction=-1)
+    assert d.conviction_leverage == pytest.approx(9.0)
+    assert d.notional_fraction == pytest.approx(1.8)
+
+
+def test_max_leverage_parameterization_is_live_not_a_default():
+    # The drift guard: the kernel default is 5.0, blue_parity passes 8.0, live is
+    # 9.0. Passing different caps MUST change the saturated conviction — proving
+    # VIOLET never inherits a hardcoded/drifted value.
+    cap5 = _sizer(5.0).calculate(capital=69_000.0, vel_div=-0.20, trade_direction=-1)
+    cap9 = _sizer(9.0).calculate(capital=69_000.0, vel_div=-0.20, trade_direction=-1)
+    assert cap5.conviction_leverage == pytest.approx(5.0)
+    assert cap9.conviction_leverage == pytest.approx(9.0)
+    assert cap5.conviction_leverage != cap9.conviction_leverage
+
+
+def test_size_decision_is_typed_and_frozen():
+    d = _sizer().calculate(capital=50_000.0, vel_div=-0.04, trade_direction=-1)
+    assert isinstance(d, SizeDecision)
+    with pytest.raises(ValidationError):
+        d.fraction = 0.5  # frozen
+
+
+# ── selection returns a typed pick or None ────────────────────────────────────
+
+def test_selector_returns_typed_short_pick():
+    pick = VioletAssetSelector(lookback_horizon=50).pick(_universe(), regime_direction=-1)
+    if pick is not None:
+        assert isinstance(pick, AssetPick)
+        assert pick.side == "SHORT"
+
+
+def test_selector_empty_universe_is_none():
+    assert VioletAssetSelector(lookback_horizon=50).pick({}, regime_direction=-1) is None
+
+
+# ── property: conviction always within the kernel envelope; output finite ─────
+
+@given(
+    vel_div=st.floats(min_value=-0.5, max_value=0.5, allow_nan=False, allow_infinity=False),
+    cap=st.floats(min_value=1.0, max_value=1e7, allow_nan=False, allow_infinity=False),
+)
+@settings(max_examples=60, deadline=None)
+def test_conviction_within_envelope_and_finite(vel_div, cap):
+    d = _sizer(9.0).calculate(capital=cap, vel_div=vel_div, trade_direction=-1)
+    assert 0.0 <= d.conviction_leverage <= 9.0 + 1e-9
+    assert math.isfinite(d.notional_fraction) and d.notional_fraction >= 0.0
+    assert 0 <= d.bucket_idx <= 3