malkhut(tests): 1140 test functions across 46 test files
CWM (103): core mechanics, exhaustive edge cases, numba, exchange mechanics
Replay (118): exhaustive verification, microstructure, trajectory
Training (190): asset classification, phase0 extensive, pipeline, exhaustive
DSL (102): v2 syntax, expanded, new features
ASEx (33): validate-before-mutate, single-writer
Planner (48): MCTS, alternatives, hooks
Counterparties (19): 9 adversarial agent policies
Clock (30): event-driven reactor
BingX (28): venue adapter
IPC (8): Zinc SHM
Storage (9): ClickHouse
Risk (4): hard invariants
State (17): frozen dataclass invariants
Integration: E2E, concurrency, sync/async seams, hypothesis, fuzz, adversarial
2026-07-11 10:46:12 +02:00
|
|
|
"""
|
|
|
|
|
Adversarial scenario tests.
|
|
|
|
|
|
|
|
|
|
These prove the core thesis: mixed policies survive diverse counterparty
|
|
|
|
|
ecologies better than pure deterministic quotes.
|
|
|
|
|
"""
|
|
|
|
|
import pytest
|
|
|
|
|
from malkhut.state import (
|
|
|
|
|
AccountState, ExecutionIntent, FulfilmentPolicyParams, IntentKind,
|
|
|
|
|
MarketWorldState, Mode, OrderBookState, PriceLevel, Side, VenueRules,
|
|
|
|
|
)
|
|
|
|
|
from malkhut.cwm.core import MinimalCryptoLOBCWM
|
|
|
|
|
from malkhut.planner.sm_mcts import DecoupledUCBPlanner
|
|
|
|
|
from malkhut.planner.action_menu import build_our_actions
|
|
|
|
|
from malkhut.risk.gate import RiskGate
|
|
|
|
|
from malkhut.counterparties import ToxicTakerPolicy, default_counterparty_ecology
|
|
|
|
|
from malkhut.actions import ActionKind, FulfilmentAction, OrderType, PlannedPolicy
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _venue():
|
|
|
|
|
return VenueRules(
|
|
|
|
|
exchange="bingx", symbol="BTCUSDT", tick_size=0.1, lot_size=0.001,
|
|
|
|
|
min_qty=0.001, min_notional=5.0, maker_fee_bps=-0.2, taker_fee_bps=0.5,
|
|
|
|
|
post_only_supported=True, reduce_only_supported=True,
|
|
|
|
|
max_orders_per_second=100, max_cancels_per_minute=120,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _params(**kw):
|
|
|
|
|
d = dict(
|
|
|
|
|
version="adv", ucb_c=1.414, max_sims=64, max_depth=2,
|
|
|
|
|
rollout_depth=2, root_temperature=0.5, min_root_entropy=0.25,
|
|
|
|
|
quote_offsets_ticks=(0, 1, 2), quote_size_fractions=(0.10, 0.25, 0.50),
|
|
|
|
|
passive_ttl_ms=200, aggressive_ttl_ms=50,
|
|
|
|
|
maker_edge_min_bps=0.5, cross_spread_edge_min_bps=5.0,
|
|
|
|
|
adverse_toxicity_cancel_threshold=0.5, queue_churn_cancel_threshold=0.5,
|
|
|
|
|
mae_tail_cut_bps=50.0, mfe_giveback_cut_fraction=0.5,
|
|
|
|
|
max_time_in_loss_s=300.0, failed_recovery_cut_count=3,
|
|
|
|
|
recovery_velocity_min_bps_per_s=0.0,
|
|
|
|
|
max_symbol_notional_fraction=0.20, max_single_order_notional_fraction=0.05,
|
|
|
|
|
reduce_when_global_up_fraction=0.30, session_profit_lock_fraction=0.02,
|
|
|
|
|
w_expected_pnl=1.0, w_fill_probability=0.5, w_adverse_selection=2.0,
|
|
|
|
|
w_queue_priority=0.5, w_inventory_risk=1.5, w_tail_loss=5.0,
|
|
|
|
|
w_fee_quality=0.5, w_time_decay=0.3, w_policy_entropy=0.5,
|
|
|
|
|
robust_tail_weight=2.0, toxic_counterparty_weight=3.0,
|
|
|
|
|
low_liquidity_weight=2.0, latency_stress_weight=1.0,
|
|
|
|
|
)
|
|
|
|
|
d.update(kw)
|
|
|
|
|
return FulfilmentPolicyParams(**d)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def _state_with_intent(**kw):
|
|
|
|
|
from malkhut.state import TradePathState, AccountState as AC
|
|
|
|
|
tp = kw.get("trade_path")
|
|
|
|
|
return MarketWorldState(
|
|
|
|
|
ts_ns=1_000_000_000, mode=Mode.REPLAY_NO_IMPACT, venue=_venue(),
|
|
|
|
|
book=OrderBookState(
|
|
|
|
|
ts_ns=1_000_000_000, symbol="BTCUSDT",
|
|
|
|
|
bids=(PriceLevel(50000.0, 1.0), PriceLevel(49999.0, 2.0)),
|
|
|
|
|
asks=(PriceLevel(50001.0, 1.0), PriceLevel(50002.0, 2.0)),
|
|
|
|
|
),
|
|
|
|
|
account=AC(
|
|
|
|
|
ts_ns=1_000_000_000, equity=10000.0, wallet_balance=10000.0,
|
|
|
|
|
available_balance=10000.0, margin_used=0.0, total_notional=0.0,
|
|
|
|
|
),
|
|
|
|
|
intent=ExecutionIntent(
|
|
|
|
|
intent_id="adv", ts_ns=1_000_000_000, symbol="BTCUSDT",
|
|
|
|
|
kind=IntentKind.ENTER_LONG, target_qty=0.01, max_notional=500.0,
|
|
|
|
|
urgency=0.5, alpha_horizon_s=60.0, alpha_bps=2.0,
|
|
|
|
|
max_slippage_bps=5.0, prefer_maker=True, reduce_only=False,
|
|
|
|
|
ttl_s=300.0, reason="adversarial_test",
|
|
|
|
|
),
|
|
|
|
|
trade_path=tp,
|
|
|
|
|
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestToxicTakerPicksOffStaleQuote:
|
|
|
|
|
def test_pure_stale_quote_vulnerable(self):
|
|
|
|
|
"""A pure 'always quote best bid' is predictable and gets picked off."""
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
params = _params()
|
|
|
|
|
actions = build_our_actions(state, params)
|
|
|
|
|
|
|
|
|
|
# Pure strategy: always place at best bid, 25% size
|
|
|
|
|
pure_actions = [a for a in actions if a.kind == ActionKind.PLACE and a.price_ticks_from_best == 0]
|
|
|
|
|
assert len(pure_actions) > 0
|
|
|
|
|
# This action is predictable — toxic taker can target it
|
|
|
|
|
|
|
|
|
|
def test_mixed_policy_reduces_predictability(self):
|
|
|
|
|
"""SM-MCTS should return a mixed distribution, not a single action."""
|
|
|
|
|
cwm = MinimalCryptoLOBCWM()
|
|
|
|
|
planner = DecoupledUCBPlanner(
|
|
|
|
|
cwm=cwm, counterparties=default_counterparty_ecology(), rng_seed=42,
|
|
|
|
|
)
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
params = _params()
|
|
|
|
|
result = planner.plan(root_state=state, params=params, budget_ms=15)
|
|
|
|
|
|
|
|
|
|
# Distribution should have multiple non-zero probabilities
|
|
|
|
|
nonzero = [p for p in result.probabilities if p > 0.01]
|
|
|
|
|
assert len(nonzero) >= 2, "Pure deterministic policy is exploitable"
|
|
|
|
|
|
|
|
|
|
def test_mixed_policy_includes_cancellation_option(self):
|
|
|
|
|
"""A good policy should have PASSIVE placement + NOOP as minimum diversity."""
|
|
|
|
|
cwm = MinimalCryptoLOBCWM()
|
|
|
|
|
planner = DecoupledUCBPlanner(
|
|
|
|
|
cwm=cwm, counterparties=default_counterparty_ecology(), rng_seed=42,
|
|
|
|
|
)
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
params = _params()
|
|
|
|
|
result = planner.plan(root_state=state, params=params, budget_ms=15)
|
|
|
|
|
|
|
|
|
|
# Action set should include NOOP and at least one passive placement
|
|
|
|
|
all_kinds = set(a.kind for a in result.actions)
|
|
|
|
|
assert ActionKind.NOOP in all_kinds
|
|
|
|
|
assert ActionKind.PLACE in all_kinds
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestRiskGateAdversarial:
|
|
|
|
|
def test_kill_switch_blocks_all(self):
|
|
|
|
|
gate = RiskGate()
|
|
|
|
|
gate._kill_switch_active = lambda: True
|
|
|
|
|
action = FulfilmentAction(ActionKind.PLACE, Side.BUY, OrderType.LIMIT, 0, 0.1, 200)
|
|
|
|
|
planned = PlannedPolicy(actions=(action,), probabilities=(1.0,),
|
|
|
|
|
selected_action=action, diagnostics={})
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
decision = gate.validate(state, planned, _params())
|
|
|
|
|
assert not decision.approved
|
|
|
|
|
assert decision.reason == "kill_switch"
|
|
|
|
|
|
|
|
|
|
def test_post_only_cross_rejected(self):
|
|
|
|
|
gate = RiskGate()
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
action = FulfilmentAction(
|
malkhut(wire): OrderType as three orthogonal dimensions — Fable's corrections
CRITICAL REFACTOR based on Fable's review (S9 roadmap item):
Before: flat enum conflating order types with TIF/instructions
OrderType had MARKET, LIMIT, IOC, FOK, POST_ONLY, REDUCE_ONLY, etc.
After: three orthogonal dimensions (FIX-aligned):
1. OrderType (Tag 40): what the order IS
LIMIT, MARKET, STOP_MARKET, STOP_LIMIT, TRIGGER_MARKET, TRIGGER_LIMIT,
TRAILING_STOP, OCO, TP_SL
2. TimeInForce (Tag 59): how long it LIVES
GTC, IOC, FOK, GTD
3. Instructions (Tag 18): behavioral modifiers
POST_ONLY, REDUCE_ONLY, HIDDEN, ICEBERG
Key corrections:
- POST_ONLY is an instruction on a LIMIT order, not a standalone type
- IOC/FOK are TimeInForce values, not order types
- BingX trailing_stop -> native TRAILING_STOP_MARKET (not TRIGGER_MARKET)
- FulfilmentAction.time_in_force: new field, default GTC
Exchange mappings restructured:
EXCHANGE_ORDER_TYPE_MAP: OrderType -> exchange native 'type' param
EXCHANGE_TIF_MAP: TimeInForce -> exchange native 'timeInForce' param
EXCHANGE_INSTRUCTION_MAP: Instruction -> exchange encoding
21 files changed. 380+ tests pass. Backward compatible.
2026-07-14 14:46:44 +02:00
|
|
|
ActionKind.PLACE, Side.BUY, OrderType.LIMIT, -10, 0.1, 200, post_only=True,
|
malkhut(tests): 1140 test functions across 46 test files
CWM (103): core mechanics, exhaustive edge cases, numba, exchange mechanics
Replay (118): exhaustive verification, microstructure, trajectory
Training (190): asset classification, phase0 extensive, pipeline, exhaustive
DSL (102): v2 syntax, expanded, new features
ASEx (33): validate-before-mutate, single-writer
Planner (48): MCTS, alternatives, hooks
Counterparties (19): 9 adversarial agent policies
Clock (30): event-driven reactor
BingX (28): venue adapter
IPC (8): Zinc SHM
Storage (9): ClickHouse
Risk (4): hard invariants
State (17): frozen dataclass invariants
Integration: E2E, concurrency, sync/async seams, hypothesis, fuzz, adversarial
2026-07-11 10:46:12 +02:00
|
|
|
)
|
|
|
|
|
planned = PlannedPolicy(actions=(action,), probabilities=(1.0,),
|
|
|
|
|
selected_action=action, diagnostics={})
|
|
|
|
|
decision = gate.validate(state, planned, _params())
|
|
|
|
|
assert not decision.approved
|
|
|
|
|
|
|
|
|
|
def test_leverage_exceeded_blocks(self):
|
|
|
|
|
gate = RiskGate()
|
|
|
|
|
from malkhut.state import AccountState
|
|
|
|
|
state = MarketWorldState(
|
|
|
|
|
ts_ns=1_000_000_000, mode=Mode.LIVE, venue=_venue(),
|
|
|
|
|
book=OrderBookState(ts_ns=1, symbol="BTCUSDT",
|
|
|
|
|
bids=(PriceLevel(50000.0, 1.0),),
|
|
|
|
|
asks=(PriceLevel(50001.0, 1.0),)),
|
|
|
|
|
account=AccountState(
|
|
|
|
|
ts_ns=1, equity=1000.0, wallet_balance=1000.0,
|
|
|
|
|
available_balance=1000.0, margin_used=0.0, total_notional=5000.0,
|
|
|
|
|
),
|
|
|
|
|
)
|
|
|
|
|
action = FulfilmentAction(ActionKind.PLACE, Side.BUY, OrderType.LIMIT, 0, 0.1, 200)
|
|
|
|
|
planned = PlannedPolicy(actions=(action,), probabilities=(1.0,),
|
|
|
|
|
selected_action=action, diagnostics={})
|
|
|
|
|
decision = gate.validate(state, planned, _params())
|
|
|
|
|
assert not decision.approved
|
|
|
|
|
assert decision.reason == "leverage_limit"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestCounterpartyAdversarial:
|
|
|
|
|
def test_toxic_taker_attacks_high_toxicity(self):
|
|
|
|
|
"""When orderflow toxicity is high, toxic taker should cross."""
|
|
|
|
|
from malkhut.state import TradePathState
|
|
|
|
|
tp = TradePathState(
|
|
|
|
|
symbol="BTCUSDT", side=Side.BUY, entry_ts_ns=0, now_ts_ns=100_000_000,
|
|
|
|
|
bars_held=10, seconds_held=100.0, pnl_bps=0.0, mae_bps=-10.0,
|
|
|
|
|
mfe_bps=15.0, distance_from_mfe_bps=15.0, distance_from_entry_bps=0.0,
|
|
|
|
|
time_to_mfe_s=30.0, time_in_loss_s=50.0, time_in_profit_s=50.0,
|
|
|
|
|
time_since_last_profit_s=10.0, time_since_deep_mae_s=20.0,
|
|
|
|
|
loss_to_profit_transitions=1, deep_loss_recoveries=0,
|
|
|
|
|
failed_recovery_count=0, recovery_velocity_bps_per_s=1.0,
|
|
|
|
|
adverse_velocity_bps_per_s=-0.5,
|
|
|
|
|
dolphin_regime_score=0.5, jericho_signal_strength=0.3,
|
|
|
|
|
volatility_bps=15.0, orderflow_toxicity=0.9,
|
|
|
|
|
queue_churn_score=0.2, book_imbalance=0.1, cross_venue_lead_score=0.1,
|
|
|
|
|
)
|
|
|
|
|
state = _state_with_intent(trade_path=tp)
|
|
|
|
|
toxic = ToxicTakerPolicy()
|
|
|
|
|
import random
|
|
|
|
|
action = toxic.rollout_action(state, random.Random(42))
|
|
|
|
|
assert action.kind == ActionKind.CROSS_SPREAD
|
|
|
|
|
|
|
|
|
|
def test_latency_arb_attacks_stale_quotes(self):
|
|
|
|
|
"""Latency arb crosses when cross-venue lead is strong."""
|
|
|
|
|
from malkhut.state import TradePathState
|
|
|
|
|
from malkhut.counterparties import LatencyArbPolicy
|
|
|
|
|
tp = TradePathState(
|
|
|
|
|
symbol="BTCUSDT", side=Side.BUY, entry_ts_ns=0, now_ts_ns=100_000_000,
|
|
|
|
|
bars_held=10, seconds_held=100.0, pnl_bps=0.0, mae_bps=-10.0,
|
|
|
|
|
mfe_bps=15.0, distance_from_mfe_bps=15.0, distance_from_entry_bps=0.0,
|
|
|
|
|
time_to_mfe_s=30.0, time_in_loss_s=50.0, time_in_profit_s=50.0,
|
|
|
|
|
time_since_last_profit_s=10.0, time_since_deep_mae_s=20.0,
|
|
|
|
|
loss_to_profit_transitions=1, deep_loss_recoveries=0,
|
|
|
|
|
failed_recovery_count=0, recovery_velocity_bps_per_s=1.0,
|
|
|
|
|
adverse_velocity_bps_per_s=-0.5,
|
|
|
|
|
dolphin_regime_score=0.5, jericho_signal_strength=0.3,
|
|
|
|
|
volatility_bps=15.0, orderflow_toxicity=0.3,
|
|
|
|
|
queue_churn_score=0.2, book_imbalance=0.1, cross_venue_lead_score=0.9,
|
|
|
|
|
)
|
|
|
|
|
state = _state_with_intent(trade_path=tp)
|
|
|
|
|
arb = LatencyArbPolicy()
|
|
|
|
|
import random
|
|
|
|
|
action = arb.rollout_action(state, random.Random(42))
|
|
|
|
|
assert action.kind == ActionKind.CROSS_SPREAD
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class TestMixedPolicySurvivesEcology:
|
|
|
|
|
def test_noop_always_available(self):
|
|
|
|
|
"""NOOP must always be in the action set — sometimes the best quote is no quote."""
|
|
|
|
|
state = _state_with_intent()
|
|
|
|
|
params = _params()
|
|
|
|
|
actions = build_our_actions(state, params)
|
|
|
|
|
kinds = [a.kind for a in actions]
|
|
|
|
|
assert ActionKind.NOOP in kinds
|
|
|
|
|
|
|
|
|
|
def test_exit_available_under_tail_risk(self):
|
|
|
|
|
"""When path risk is high, FULL_EXIT must be available."""
|
|
|
|
|
from malkhut.state import TradePathState
|
|
|
|
|
tp = TradePathState(
|
|
|
|
|
symbol="BTCUSDT", side=Side.BUY, entry_ts_ns=0, now_ts_ns=100_000_000,
|
|
|
|
|
bars_held=10, seconds_held=100.0, pnl_bps=-30.0, mae_bps=-60.0,
|
|
|
|
|
mfe_bps=5.0, distance_from_mfe_bps=35.0, distance_from_entry_bps=30.0,
|
|
|
|
|
time_to_mfe_s=10.0, time_in_loss_s=90.0, time_in_profit_s=10.0,
|
|
|
|
|
time_since_last_profit_s=80.0, time_since_deep_mae_s=5.0,
|
|
|
|
|
loss_to_profit_transitions=0, deep_loss_recoveries=0,
|
|
|
|
|
failed_recovery_count=4, recovery_velocity_bps_per_s=-2.0,
|
|
|
|
|
adverse_velocity_bps_per_s=3.0,
|
|
|
|
|
dolphin_regime_score=0.2, jericho_signal_strength=0.1,
|
|
|
|
|
volatility_bps=30.0, orderflow_toxicity=0.7,
|
|
|
|
|
queue_churn_score=0.5, book_imbalance=0.3, cross_venue_lead_score=-0.5,
|
|
|
|
|
)
|
|
|
|
|
state = _state_with_intent(trade_path=tp)
|
|
|
|
|
params = _params()
|
|
|
|
|
actions = build_our_actions(state, params)
|
|
|
|
|
kinds = [a.kind for a in actions]
|
|
|
|
|
assert ActionKind.FULL_EXIT in kinds
|