How "Smart Money" Hunts Retail: Trading Liquidation Maps in Practice
A liquidation heatmap is essentially an X-ray of retail hope and greed. While retail traders set their stop-losses based on classic TA right behind local high/low levels, institutional players see those exact clusters as prime exit liquidity for their own positions.
Market makers simply cannot fill or exit multi-million dollar positions in an illiquid order book without massive slippage. They need raw liquidity—and forced margin calls provide it on a silver platter.
The Short Trap: Mechanics of a Classic Short Squeeze
Picture this: an asset has been dumping for three days straight. The herd smells blood, stacks up 20x–50x leveraged shorts, and tucks their stop-losses (or liquidation points) right above local resistance—usually right around a clean psychological round number or the top of a consolidation range.
What does a whale do when they need to build a massive long or offload heavy bags?
- The Slow Grind: Price slowly creeps upward without triggering alarm bells, grinding right up against the liquidation cluster on the heatmap.
- The Impulse Spike: One violent market buy pushes price 1.5–2% past key resistance.
- The Cascade: Stops detonate, and high-leverage accounts get instantly wiped out. The exchange engine slams forced market buy orders to close those short positions.
- Volume Spike: Panic market buys from exchange margin calls smash directly into the whale’s pre-placed limit sell orders, filling their massive ask depth effortlessly.
Retail trader thinks: "Breakout! Trend reversal confirmed, apeing long!" They buy right at the peak of the impulse, the market maker flips their position, and price dumps like a stone to hunt the long stops down below.
Why Order Books Get Blown Out
When a whale with a $500,000 position on 20x leverage gets liquidated, the exchange fires off a massive market order to close it out. If there isn't enough limit liquidity sitting on the opposite side of the book, price slices straight through levels.
- Chewing Through Limits: The liquidation engine tears through the order book. If there are only 5 BTC worth of limit orders at $60,000 and the engine needs to dump 20 BTC, it blows right past that level until it finds enough volume.
- The Domino Effect: Slipping through levels triggers adjacent stop-losses and cascades down to lower-leverage positions (like traders sitting on 10x).
- The V-Bottom (Dead Cat / Instant Reversal): Once the cascade exhausts all available liquidity on the move, the order book is left completely hollow. That’s when the market maker step in with heavy limit orders, absorbing the residual noise and instantly snapping price back the other way.
Playbook Setups: Extracting Value from Liquidation Zones
Don't try to trade the exact moment a liquidation hit in real-time—you'll get wrecked by wide spreads and slippage. Play the market's reaction *after* the initial fire dies down.
Setup #1: The Liquidation Sweep (Stop-Hunt Reversal)
- Conditions: The heatmap shows a massive overhead or underlying cluster of shorts/longs less than 1% away from current price.
- Game Plan: Wait for the level to get swept. Do NOT trade the initial breakout spike.
- Execution: Price aggressively pierces the key level, 1-minute volume goes parabolic, and Open Interest (OI) plummets (position reset). Once the candle closes leaving a long wick/tail back inside the range—fade the move and enter in the opposite direction.
- Stop-Loss: Place it strictly past the high/low wick of the sweep candle. If price pushes past that wick again, it’s a real directional breakout, not a liquidity sweep. Keep your risk tight.
Automated Tracking: Anomaly Detection Script
To avoid staring at charts 24/7 waiting for a liquidation cascade, run the script below. It monitors the real-time exchange liquidation feed and prints outlier liquidation sizes right to your terminal.
Written in pure Python with zero external dependencies beyond standard networking, built to run continuously without crashing on connection drops.
import json
import time
import threading
import logging
from collections import deque, OrderedDict
from statistics import median
from typing import Dict, Tuple
import requests
import websocket
from requests.adapters import HTTPAdapter
from urllib3.util.retry import Retry
class ProductionMarketEngine:
"""
Production-grade Binance Futures liquidation / market-state engine.
Data sources:
- Binance Futures forceOrder WebSocket
- Binance Futures ticker WebSocket
- Binance Futures Open Interest REST
Main analysis window:
5 minutes
Features:
- Rolling liquidation window
- Calendar-aligned 5-minute liquidation buckets
- TTL event deduplication
- Real market price feed
- Exact 5-minute OI delta
- Liquidation imbalance
- Liquidation intensity
- Price momentum
- Volatility
- Market-state classification
- Signal scoring
- WebSocket reconnect with exponential backoff
- HTTP retries for Open Interest
- Graceful shutdown
- Thread-safe state
- Runtime health monitoring
"""
def __init__(
self,
symbol: str = "btcusdt",
window_seconds: int = 300,
oi_poll_seconds: int = 10,
baseline_buckets: int = 288,
min_baseline_buckets: int = 12,
):
self.symbol = symbol.lower()
self.symbol_upper = symbol.upper()
self.window_seconds = window_seconds
self.oi_poll_seconds = oi_poll_seconds
# 288 × 5 minutes = 24 hours.
self.baseline_buckets = baseline_buckets
# Minimum completed buckets before baseline becomes reliable.
self.min_baseline_buckets = min_baseline_buckets
# --------------------------------------------------------------
# Thread synchronization
# --------------------------------------------------------------
self.lock = threading.RLock()
# Event used for interruptible thread shutdown / waiting.
self.stop_event = threading.Event()
# --------------------------------------------------------------
# Runtime state
# --------------------------------------------------------------
self.is_running = False
self.ws_combined = None
self.ws_thread = None
self.oi_thread = None
self.maintenance_thread = None
# --------------------------------------------------------------
# Liquidation rolling window
# --------------------------------------------------------------
self.events = deque()
# event_id -> received timestamp
self.seen_ids = OrderedDict()
self.dedup_ttl_seconds = max(
window_seconds * 2,
600,
)
# --------------------------------------------------------------
# Real market price history
# --------------------------------------------------------------
self.price_history = deque()
self.current_price = 0.0
self.last_price_timestamp = 0.0
# --------------------------------------------------------------
# Open Interest
# --------------------------------------------------------------
self.oi_history = deque(
maxlen=180
)
self.current_oi = 0.0
self.last_oi_timestamp = 0.0
# --------------------------------------------------------------
# Calendar-aligned liquidation buckets
# --------------------------------------------------------------
# {
# bucket_timestamp: liquidation_notional
# }
self.bucket_accum = OrderedDict()
# --------------------------------------------------------------
# WebSocket state
# --------------------------------------------------------------
self.ws_connected = False
self.last_ws_message = 0.0
self.ws_reconnects = 0
# --------------------------------------------------------------
# Diagnostics
# --------------------------------------------------------------
self.total_liquidation_events = 0
self.invalid_liquidation_events = 0
# --------------------------------------------------------------
# Logger
# --------------------------------------------------------------
self.logger = logging.getLogger(
f"ProductionMarketEngine.{self.symbol_upper}"
)
# ==================================================================
# TIME / BUCKET HELPERS
# ==================================================================
@staticmethod
def _bucket_timestamp(timestamp: float) -> int:
"""
Convert timestamp to calendar-aligned 5-minute bucket.
"""
return int(timestamp // 300) * 300
# ==================================================================
# CLEANUP
# ==================================================================
def _clean_ttl_cache(self, now: float) -> None:
"""
Remove expired liquidation event IDs.
"""
cutoff = now - self.dedup_ttl_seconds
while self.seen_ids:
first_key, first_timestamp = next(
iter(self.seen_ids.items())
)
if first_timestamp < cutoff:
self.seen_ids.popitem(
last=False
)
else:
break
def _clean_old_events(self, now: float) -> None:
"""
Keep liquidation and price data only inside
the rolling analysis window.
"""
cutoff = now - self.window_seconds
while self.events:
if self.events[0]["timestamp"] < cutoff:
self.events.popleft()
else:
break
while self.price_history:
if self.price_history[0]["timestamp"] < cutoff:
self.price_history.popleft()
else:
break
def _clean_old_buckets(self, now: float) -> None:
"""
Remove old liquidation buckets.
This method is called from the dedicated maintenance thread,
so bucket cleanup does NOT depend on new liquidation events.
"""
current_bucket = self._bucket_timestamp(now)
minimum_bucket = (
current_bucket
- (self.baseline_buckets + 2) * 300
)
while self.bucket_accum:
first_bucket = next(
iter(self.bucket_accum)
)
if first_bucket < minimum_bucket:
self.bucket_accum.popitem(
last=False
)
else:
break
# ==================================================================
# LIQUIDATION INGESTION
# ==================================================================
def _process_liquidation(
self,
order_data: dict,
) -> None:
"""
Process Binance Futures forceOrder event.
"""
if not isinstance(
order_data,
dict,
):
return
now = time.time()
raw_side = str(
order_data.get(
"S",
"",
)
).upper()
# --------------------------------------------------------------
# Validate side before touching deduplication cache.
# --------------------------------------------------------------
if raw_side == "SELL":
side = "LONG_LIQ"
elif raw_side == "BUY":
side = "SHORT_LIQ"
else:
with self.lock:
self.invalid_liquidation_events += 1
return
try:
execution_timestamp_ms = int(
order_data.get(
"T",
int(now * 1000),
)
)
event_timestamp = (
execution_timestamp_ms
/ 1000.0
)
price = float(
order_data.get(
"ap",
order_data.get(
"p",
0,
),
)
)
qty = float(
order_data.get(
"q",
0,
)
)
order_id = str(
order_data.get(
"i",
"",
)
)
except (
TypeError,
ValueError,
):
with self.lock:
self.invalid_liquidation_events += 1
self.logger.warning(
"Invalid liquidation event received."
)
return
if price <= 0 or qty <= 0:
with self.lock:
self.invalid_liquidation_events += 1
return
notional = price * qty
if notional <= 0:
return
# --------------------------------------------------------------
# Event ID
#
# Prefer Binance order ID when available.
# If unavailable, use deterministic composite fallback.
# --------------------------------------------------------------
if order_id:
event_id = (
f"{execution_timestamp_ms}_"
f"{order_id}"
)
else:
event_id = (
f"{execution_timestamp_ms}_"
f"{side}_"
f"{price:.12f}_"
f"{qty:.12f}"
)
event = {
"id": event_id,
"timestamp": event_timestamp,
"side": side,
"price": price,
"qty": qty,
"notional": notional,
}
with self.lock:
self._clean_ttl_cache(
now
)
if event_id in self.seen_ids:
return
self.seen_ids[event_id] = now
self.events.append(
event
)
# Calendar-aligned bucket.
bucket = self._bucket_timestamp(
event_timestamp
)
if bucket not in self.bucket_accum:
self.bucket_accum[bucket] = 0.0
self.bucket_accum[bucket] += (
notional
)
self.total_liquidation_events += 1
# ==================================================================
# MARKET PRICE INGESTION
# ==================================================================
def _process_ticker(
self,
ticker_data: dict,
) -> None:
"""
Process Binance ticker event.
"""
try:
price = float(
ticker_data.get(
"c",
0,
)
)
except (
TypeError,
ValueError,
):
return
if price <= 0:
return
now = time.time()
with self.lock:
self.current_price = price
self.last_price_timestamp = now
self.price_history.append(
{
"timestamp": now,
"price": price,
}
)
# ==================================================================
# OPEN INTEREST HTTP SESSION
# ==================================================================
@staticmethod
def _create_http_session() -> requests.Session:
"""
Create requests Session with connection pooling
and retry policy.
"""
session = requests.Session()
retry = Retry(
total=4,
connect=4,
read=4,
status=4,
backoff_factor=0.5,
status_forcelist=(
429,
500,
502,
503,
504,
),
allowed_methods=frozenset(
[
"GET",
]
),
respect_retry_after_header=True,
)
adapter = HTTPAdapter(
max_retries=retry,
pool_connections=4,
pool_maxsize=4,
)
session.mount(
"https://",
adapter,
)
session.headers.update(
{
"User-Agent":
"ProductionMarketEngine/1.0"
}
)
return session
# ==================================================================
# OPEN INTEREST POLLER
# ==================================================================
def _poll_open_interest(
self,
) -> None:
"""
Poll Binance Futures Open Interest.
"""
url = (
"https://fapi.binance.com"
"/fapi/v1/openInterest"
)
session = (
self._create_http_session()
)
try:
while not self.stop_event.is_set():
try:
response = session.get(
url,
params={
"symbol":
self.symbol_upper
},
timeout=5,
)
response.raise_for_status()
data = response.json()
oi = float(
data.get(
"openInterest",
0,
)
)
if oi <= 0:
raise ValueError(
"Invalid Open Interest."
)
now = time.time()
with self.lock:
self.current_oi = oi
self.last_oi_timestamp = now
self.oi_history.append(
{
"timestamp": now,
"oi": oi,
}
)
except requests.RequestException as exc:
self.logger.warning(
"Open Interest request failed: %s",
exc,
)
except (
ValueError,
TypeError,
) as exc:
self.logger.warning(
"Invalid Open Interest response: %s",
exc,
)
except Exception:
self.logger.exception(
"Unexpected Open Interest error."
)
# Interruptible wait.
self.stop_event.wait(
self.oi_poll_seconds
)
finally:
session.close()
# ==================================================================
# MAINTENANCE THREAD
# ==================================================================
def _maintenance_loop(
self,
) -> None:
"""
Periodic memory/state cleanup.
Important:
cleanup is independent of liquidation activity.
"""
while not self.stop_event.wait(10):
now = time.time()
with self.lock:
self._clean_ttl_cache(
now
)
self._clean_old_events(
now
)
self._clean_old_buckets(
now
)
# ==================================================================
# OI 5-MINUTE DELTA
# ==================================================================
def _get_oi_change_5m(
self,
now: float,
) -> Tuple[float, bool]:
"""
Calculate actual 5-minute Open Interest change.
Returns:
change_pct
data_ready
"""
if len(self.oi_history) < 2:
return 0.0, False
cutoff = (
now
- self.window_seconds
)
baseline_sample = None
for sample in self.oi_history:
if sample["timestamp"] <= cutoff:
baseline_sample = sample
else:
break
# Not enough historical data yet.
if baseline_sample is None:
oldest = self.oi_history[0]
if (
now
- oldest["timestamp"]
< self.window_seconds * 0.8
):
return 0.0, False
baseline_sample = oldest
latest_sample = (
self.oi_history[-1]
)
first_oi = baseline_sample["oi"]
last_oi = latest_sample["oi"]
if first_oi <= 0:
return 0.0, False
change_pct = (
(
last_oi
- first_oi
)
/ first_oi
) * 100.0
return change_pct, True
# ==================================================================
# PRICE METRICS
# ==================================================================
def _get_price_metrics(
self,
) -> Tuple[
float,
float,
float,
bool,
]:
"""
Returns:
current price
5m price change
5m high-low volatility
ready
"""
if len(
self.price_history
) < 2:
return (
0.0,
0.0,
0.0,
False,
)
first_price = (
self.price_history[0]["price"]
)
last_price = (
self.price_history[-1]["price"]
)
if first_price <= 0:
return (
0.0,
0.0,
0.0,
False,
)
price_change_pct = (
(
last_price
- first_price
)
/ first_price
) * 100.0
prices = [
item["price"]
for item in self.price_history
]
high_price = max(prices)
low_price = min(prices)
if low_price > 0:
volatility_pct = (
(
high_price
- low_price
)
/ low_price
) * 100.0
else:
volatility_pct = 0.0
return (
last_price,
price_change_pct,
volatility_pct,
True,
)
# ==================================================================
# LIQUIDATION METRICS
# ==================================================================
def _get_liquidation_metrics(
self,
) -> Tuple[
float,
float,
float,
float,
]:
"""
Returns:
long liquidation
short liquidation
total liquidation
imbalance
"""
long_liq = 0.0
short_liq = 0.0
for event in self.events:
if event["side"] == "LONG_LIQ":
long_liq += event["notional"]
elif event["side"] == "SHORT_LIQ":
short_liq += event["notional"]
total_liq = (
long_liq
+ short_liq
)
if total_liq > 0:
imbalance = (
(
long_liq
- short_liq
)
/ total_liq
)
else:
imbalance = 0.0
return (
long_liq,
short_liq,
total_liq,
imbalance,
)
# ==================================================================
# BASELINE
# ==================================================================
def _get_baseline(
self,
) -> Tuple[
float,
int,
bool,
]:
"""
Calculate median liquidation volume
from completed calendar-aligned 5m buckets.
Current incomplete bucket is excluded.
"""
now = time.time()
current_bucket = (
self._bucket_timestamp(
now
)
)
completed = []
for (
bucket_timestamp,
volume,
) in self.bucket_accum.items():
if (
bucket_timestamp
< current_bucket
):
completed.append(
volume
)
if not completed:
return (
0.0,
0,
False,
)
completed = completed[
-self.baseline_buckets:
]
baseline = median(
completed
)
reliable = (
len(completed)
>= self.min_baseline_buckets
)
return (
float(baseline),
len(completed),
reliable,
)
# ==================================================================
# UTILITIES
# ==================================================================
@staticmethod
def _clamp(
value: float,
minimum: float,
maximum: float,
) -> float:
return max(
minimum,
min(
maximum,
value,
),
)
# ==================================================================
# SIGNAL SCORES
# ==================================================================
def _calculate_scores(
self,
intensity: float,
imbalance: float,
price_change_pct: float,
oi_change_pct: float,
oi_ready: bool,
) -> Dict[str, float]:
"""
Calculate normalized signal-strength scores.
Scores are signal strengths, not probabilities.
"""
liquidation_pressure = (
self._clamp(
(
intensity
/ 5.0
) * 100.0,
0.0,
100.0,
)
)
imbalance_score = (
abs(imbalance)
* 100.0
)
price_impulse = (
self._clamp(
(
abs(
price_change_pct
)
/ 3.0
)
* 100.0,
0.0,
100.0,
)
)
if oi_ready:
oi_contraction = (
self._clamp(
(
abs(
min(
oi_change_pct,
0.0,
)
)
/ 3.0
)
* 100.0,
0.0,
100.0,
)
)
else:
oi_contraction = 0.0
return {
"liquidation_pressure":
liquidation_pressure,
"imbalance":
imbalance_score,
"price_impulse":
price_impulse,
"oi_contraction":
oi_contraction,
}
# ==================================================================
# MARKET ANALYSIS
# ==================================================================
def analyze_market(
self,
) -> Tuple[str, Dict]:
"""
Main market-state classifier.
"""
now = time.time()
with self.lock:
self._clean_old_events(
now
)
self._clean_ttl_cache(
now
)
self._clean_old_buckets(
now
)
(
last_price,
price_change_pct,
volatility_pct,
price_ready,
) = self._get_price_metrics()
if not price_ready:
return (
"INITIALIZING_PRICE_FEED",
{},
)
(
oi_change_pct,
oi_ready,
) = self._get_oi_change_5m(
now
)
(
long_liq,
short_liq,
total_liq,
imbalance,
) = self._get_liquidation_metrics()
(
baseline_med,
baseline_count,
baseline_ready,
) = self._get_baseline()
# ----------------------------------------------------------
# No liquidation activity.
# ----------------------------------------------------------
if total_liq <= 0:
metrics = {
"price":
last_price,
"price_change_5m":
price_change_pct,
"volatility_5m":
volatility_pct,
"oi_change_5m":
oi_change_pct,
"oi_ready":
oi_ready,
"total_liq_usd":
0.0,
"long_liq_usd":
0.0,
"short_liq_usd":
0.0,
"imbalance":
0.0,
"intensity":
0.0,
"baseline_med":
baseline_med,
"baseline_buckets":
baseline_count,
"baseline_ready":
baseline_ready,
"scores": {
"liquidation_pressure":
0.0,
"imbalance":
0.0,
"price_impulse":
self._clamp(
(
abs(
price_change_pct
)
/ 3.0
)
* 100.0,
0.0,
100.0,
),
"oi_contraction":
0.0,
},
"confidence":
0.0,
}
if (
abs(
price_change_pct
) < 0.3
):
return (
"NO_LIQUIDATION_ACTIVITY",
metrics,
)
return (
"PRICE_MOVEMENT_NO_LIQUIDATION",
metrics,
)
# ----------------------------------------------------------
# Intensity.
# ----------------------------------------------------------
if (
baseline_ready
and baseline_med > 0
):
intensity = (
total_liq
/ baseline_med
)
else:
intensity = 0.0
# ----------------------------------------------------------
# Scores.
# ----------------------------------------------------------
scores = (
self._calculate_scores(
intensity=intensity,
imbalance=imbalance,
price_change_pct=price_change_pct,
oi_change_pct=oi_change_pct,
oi_ready=oi_ready,
)
)
# ----------------------------------------------------------
# Confidence.
# ----------------------------------------------------------
confidence_components = [
scores[
"liquidation_pressure"
],
scores[
"imbalance"
],
scores[
"price_impulse"
],
]
if oi_ready:
confidence_components.append(
scores[
"oi_contraction"
]
)
confidence = (
sum(
confidence_components
)
/ len(
confidence_components
)
)
metrics = {
"price":
last_price,
"price_change_5m":
price_change_pct,
"volatility_5m":
volatility_pct,
"oi_change_5m":
oi_change_pct,
"oi_ready":
oi_ready,
"total_liq_usd":
total_liq,
"long_liq_usd":
long_liq,
"short_liq_usd":
short_liq,
"imbalance":
imbalance,
"intensity":
intensity,
"baseline_med":
baseline_med,
"baseline_buckets":
baseline_count,
"baseline_ready":
baseline_ready,
"scores":
scores,
"confidence":
confidence,
}
# ==========================================================
# CASCADES
# ==========================================================
if (
baseline_ready
and intensity >= 3.0
and imbalance >= 0.60
and price_change_pct <= -2.0
):
return (
"CASCADING_DOWNSELL",
metrics,
)
if (
baseline_ready
and intensity >= 3.0
and imbalance <= -0.60
and price_change_pct >= 2.0
):
return (
"CASCADING_PUMP",
metrics,
)
# ==========================================================
# LONG CAPITULATION
# ==========================================================
if (
baseline_ready
and oi_ready
and intensity >= 2.0
and imbalance >= 0.50
and price_change_pct <= -0.50
and oi_change_pct <= -0.50
):
return (
"LONG_CAPITULATION",
metrics,
)
# ==========================================================
# SHORT SQUEEZE EXHAUSTION
# ==========================================================
if (
baseline_ready
and oi_ready
and intensity >= 2.0
and imbalance <= -0.50
and price_change_pct >= 0.50
and oi_change_pct <= -0.50
):
return (
"SHORT_SQUEEZE_EXHAUSTION",
metrics,
)
# ==========================================================
# DELEVERAGING
# ==========================================================
if (
oi_ready
and imbalance >= 0.50
and oi_change_pct <= -0.30
):
return (
"LONG_DELEVERAGING",
metrics,
)
if (
oi_ready
and imbalance <= -0.50
and oi_change_pct <= -0.30
):
return (
"SHORT_DELEVERAGING",
metrics,
)
# ==========================================================
# EXTREME LIQUIDATION SPIKES
# ==========================================================
if (
baseline_ready
and intensity >= 2.5
and imbalance >= 0.60
):
return (
"EXTREME_LONG_LIQUIDATION_SPIKE",
metrics,
)
if (
baseline_ready
and intensity >= 2.5
and imbalance <= -0.60
):
return (
"EXTREME_SHORT_LIQUIDATION_SPIKE",
metrics,
)
# ==========================================================
# ELEVATED LIQUIDATION
# ==========================================================
if imbalance >= 0.40:
return (
"ELEVATED_LONG_LIQUIDATION",
metrics,
)
if imbalance <= -0.40:
return (
"ELEVATED_SHORT_LIQUIDATION",
metrics,
)
# ==========================================================
# BASELINE WARMUP
# ==========================================================
if not baseline_ready:
return (
"WARMING_BASELINE",
metrics,
)
if intensity >= 1.5:
return (
"BALANCED_HIGH_LIQUIDATION_VOLUME",
metrics,
)
return (
"BALANCED",
metrics,
)
# ==================================================================
# WEBSOCKET CREATION
# ==================================================================
def _create_websocket(self):
"""
Create Binance combined WebSocket.
"""
combined_stream = (
"wss://fstream.binance.com/stream"
f"?streams="
f"{self.symbol}@forceOrder/"
f"{self.symbol}@ticker"
)
def on_open(ws):
with self.lock:
self.ws_connected = True
self.logger.info(
"WebSocket connected: %s",
self.symbol_upper,
)
def on_message(
ws,
message,
):
self.last_ws_message = (
time.time()
)
try:
payload = json.loads(
message
)
stream = payload.get(
"stream",
"",
)
data = payload.get(
"data",
{},
)
if (
"forceOrder"
in stream
):
self._process_liquidation(
data.get(
"o",
{},
)
)
elif (
"ticker"
in stream
):
self._process_ticker(
data
)
except json.JSONDecodeError:
self.logger.warning(
"Invalid WebSocket JSON."
)
except Exception:
self.logger.exception(
"Unexpected WebSocket message error."
)
def on_error(
ws,
error,
):
with self.lock:
self.ws_connected = False
self.logger.warning(
"WebSocket error: %s",
error,
)
def on_close(
ws,
close_status_code,
close_msg,
):
with self.lock:
self.ws_connected = False
self.logger.warning(
"WebSocket closed: "
"code=%s message=%s",
close_status_code,
close_msg,
)
return websocket.WebSocketApp(
combined_stream,
on_open=on_open,
on_message=on_message,
on_error=on_error,
on_close=on_close,
)
# ==================================================================
# WEBSOCKET LOOP
# ==================================================================
def _websocket_loop(
self,
) -> None:
"""
Persistent WebSocket loop.
Uses exponential reconnect backoff capped at 60 seconds.
Shutdown is interruptible through stop_event.
"""
backoff = 1
while not self.stop_event.is_set():
try:
self.ws_combined = (
self._create_websocket()
)
self.ws_reconnects += 1
self.logger.info(
"Starting WebSocket "
"connection attempt #%d.",
self.ws_reconnects,
)
self.ws_combined.run_forever(
ping_interval=30,
ping_timeout=10,
skip_utf8_validation=False,
)
# If run_forever returns normally after
# an established connection, reset backoff.
if self.is_running:
backoff = 1
except Exception as exc:
self.logger.error(
"WebSocket run_forever crashed: %s",
exc,
)
finally:
with self.lock:
self.ws_connected = False
if self.stop_event.is_set():
break
self.logger.info(
"Reconnecting WebSocket in %d seconds...",
backoff,
)
# Interruptible exponential backoff.
if self.stop_event.wait(
backoff
):
break
backoff = min(
backoff * 2,
60,
)
self.logger.info(
"WebSocket loop stopped."
)
# ==================================================================
# START
# ==================================================================
def start(
self,
) -> None:
"""
Start all background threads.
"""
with self.lock:
if self.is_running:
self.logger.warning(
"Engine is already running."
)
return
self.is_running = True
self.stop_event.clear()
# --------------------------------------------------------------
# Open Interest thread
# --------------------------------------------------------------
self.oi_thread = threading.Thread(
target=self._poll_open_interest,
name=(
f"OI_Poll_"
f"{self.symbol_upper}"
),
daemon=True,
)
self.oi_thread.start()
# --------------------------------------------------------------
# WebSocket thread
# --------------------------------------------------------------
self.ws_thread = threading.Thread(
target=self._websocket_loop,
name=(
f"WS_Loop_"
f"{self.symbol_upper}"
),
daemon=True,
)
self.ws_thread.start()
# --------------------------------------------------------------
# Maintenance thread
# --------------------------------------------------------------
self.maintenance_thread = (
threading.Thread(
target=self._maintenance_loop,
name=(
f"Maintenance_"
f"{self.symbol_upper}"
),
daemon=True,
)
)
self.maintenance_thread.start()
self.logger.info(
"ProductionMarketEngine started "
"successfully: %s",
self.symbol_upper,
)
# ==================================================================
# STOP
# ==================================================================
def stop(
self,
) -> None:
"""
Graceful and interruptible shutdown.
Order:
1. Signal all workers to stop.
2. Close WebSocket.
3. Wait for threads.
4. Reset runtime state.
"""
with self.lock:
if not self.is_running:
return
self.logger.info(
"Stopping ProductionMarketEngine..."
)
self.is_running = False
self.stop_event.set()
self.ws_connected = False
# --------------------------------------------------------------
# Explicitly close WebSocket.
# This unblocks run_forever().
# --------------------------------------------------------------
ws = self.ws_combined
if ws is not None:
try:
ws.close()
except Exception as exc:
self.logger.debug(
"Error closing WebSocket: %s",
exc,
)
# --------------------------------------------------------------
# Join worker threads.
# --------------------------------------------------------------
current_thread = (
threading.current_thread()
)
threads = [
self.ws_thread,
self.oi_thread,
self.maintenance_thread,
]
for thread in threads:
if (
thread is not None
and thread.is_alive()
and thread is not current_thread
):
thread.join(
timeout=5.0
)
with self.lock:
self.ws_combined = None
self.ws_thread = None
self.oi_thread = None
self.maintenance_thread = None
self.logger.info(
"ProductionMarketEngine stopped."
)
# ==================================================================
# HEALTH
# ==================================================================
def health(
self,
) -> Dict:
"""
Runtime health information.
"""
now = time.time()
with self.lock:
ws_age = (
now
- self.last_ws_message
if self.last_ws_message > 0
else None
)
oi_age = (
now
- self.last_oi_timestamp
if self.last_oi_timestamp > 0
else None
)
price_age = (
now
- self.last_price_timestamp
if self.last_price_timestamp > 0
else None
)
return {
"running":
self.is_running,
"symbol":
self.symbol_upper,
"websocket_connected":
self.ws_connected,
"websocket_last_message_age":
ws_age,
"websocket_reconnects":
self.ws_reconnects,
"price":
self.current_price,
"price_age":
price_age,
"open_interest":
self.current_oi,
"oi_age":
oi_age,
"liquidation_events":
self.total_liquidation_events,
"invalid_liquidation_events":
self.invalid_liquidation_events,
"dedup_cache_size":
len(
self.seen_ids
),
"rolling_liquidation_events":
len(
self.events
),
"price_history_size":
len(
self.price_history
),
"oi_history_size":
len(
self.oi_history
),
"baseline_buckets":
len(
self.bucket_accum
),
}
# ======================================================================
# CONSOLE OUTPUT
# ======================================================================
def print_market_state(
state: str,
metrics: Dict,
) -> None:
"""
Human-readable realtime market state.
"""
if not metrics:
print(
f"\r[{time.strftime('%H:%M:%S')}] "
f"Инициализация потоков...",
end="",
flush=True,
)
return
scores = metrics.get(
"scores",
{},
)
print(
f"\n[{time.strftime('%H:%M:%S')}] "
f"СТАТУС: >>> {state} <<<"
)
print(
f" ├─ Цена BTC: "
f"${metrics['price']:,.2f}"
)
print(
f" ├─ Цена 5M: "
f"{metrics['price_change_5m']:+.2f}%"
)
print(
f" ├─ Волатильность 5M: "
f"{metrics['volatility_5m']:.2f}%"
)
oi_status = (
"READY"
if metrics["oi_ready"]
else "WARMING"
)
print(
f" ├─ Open Interest 5M: "
f"{metrics['oi_change_5m']:+.2f}% "
f"[{oi_status}]"
)
print(
f" ├─ Ликвидации 5M: "
f"${metrics['total_liq_usd']:,.0f}"
)
print(
f" │ ├─ Long: "
f"${metrics['long_liq_usd']:,.0f}"
)
print(
f" │ └─ Short: "
f"${metrics['short_liq_usd']:,.0f}"
)
print(
f" ├─ Imbalance: "
f"{metrics['imbalance']:+.3f}"
)
print(
f" ├─ Intensity: "
f"{metrics['intensity']:.2f}x"
)
print(
f" ├─ Baseline median: "
f"${metrics['baseline_med']:,.0f}"
)
print(
f" ├─ Baseline buckets: "
f"{metrics['baseline_buckets']}"
f"{' [READY]' if metrics['baseline_ready'] else ' [WARMING]'}"
)
print(
f" ├─ Liquidation score: "
f"{scores.get('liquidation_pressure', 0):.0f}/100"
)
print(
f" ├─ Imbalance score: "
f"{scores.get('imbalance', 0):.0f}/100"
)
print(
f" ├─ Price impulse: "
f"{scores.get('price_impulse', 0):.0f}/100"
)
print(
f" ├─ OI contraction: "
f"{scores.get('oi_contraction', 0):.0f}/100"
)
print(
f" └─ Confidence: "
f"{metrics.get('confidence', 0):.0f}/100"
)
# ======================================================================
# LOGGING
# ======================================================================
def configure_logging() -> None:
logging.basicConfig(
level=logging.INFO,
format=(
"%(asctime)s | "
"%(levelname)s | "
"%(name)s | "
"%(message)s"
),
datefmt="%Y-%m-%d %H:%M:%S",
)
# ======================================================================
# APPLICATION
# ======================================================================
if __name__ == "__main__":
configure_logging()
engine = ProductionMarketEngine(
symbol="btcusdt",
window_seconds=300,
oi_poll_seconds=10,
baseline_buckets=288,
min_baseline_buckets=12,
)
print(
"Starting Production Market Engine..."
)
print(
"Streams:"
)
print(
" - Binance Futures forceOrder"
)
print(
" - Binance Futures ticker"
)
print(
" - Binance Futures Open Interest"
)
print(
"Analysis window: 5 minutes"
)
print(
"Baseline: 288 calendar-aligned 5M buckets"
)
print(
"Baseline warm-up: "
f"{engine.min_baseline_buckets} buckets"
)
engine.start()
try:
while True:
state, metrics = (
engine.analyze_market()
)
print_market_state(
state,
metrics,
)
time.sleep(10)
except KeyboardInterrupt:
print(
"\n\nStopping engine..."
)
finally:
engine.stop()
print(
"Engine stopped."
)Integrating This Into Your Workflow:
- Adjust `threshold_usd` to match your account sizing and target asset. For major caps ($BTC, $ETH), setting a $500k+ threshold filters out noise and highlights actual institutional stop sweeps.
- When the alert triggers, flip straight over to your order flow / cluster charts (like EXMON). If a massive forced liquidation print aligns with a major support or resistance level, that's your primary signal to stage a mean-reversion counter-trade.
Why Heatmaps Lie to Solo Traders (And How to Pair Them with Funding Rates)
The single biggest rookie mistake is looking at liquidation heatmaps in a vacuum. A dense cluster of red or green levels on your chart doesn't guarantee a reversal. Heatmaps show potential fuel, but they don't tell you where the main engine actually wants to drive the market.
To filter out fake-outs from true, high-conviction sweeps, you need to cross-reference the Funding Rate with Open Interest (OI):
- Sky-high Positive Funding + Surging Open Interest: The herd is aggressively piling into leveraged longs. Everyone is hyper-bullish. This creates the absolute perfect storm for a brutal long squeeze to wipe the floor clean. Expect the lower liquidation map to be heavily stacked.
- Deeply Negative Funding + Surging Open Interest: The market is completely choked with shorts. Any minor upward impulse threatens a cascading short squeeze, as forced buy-backs get triggered to cover those borrowed positions.
If you catch a massive cluster of short liquidations on the heatmap while funding has been pinned deep in the red for hours, the odds of a violent upward sweep skyrocket. Whales and market makers simply won't leave that juicy liquidity on the table.
Setup #2: The True Sweep Breakout (Trend Continuation)
Not every liquidation hunt results in an instant V-shaped rejection. Sometimes the accumulated fuel is so massive that instead of leaving a quick wick, the price ignites a powerful, multi-hour momentum trend.
Here's how to spot a real breakout through a liquidation zone:
- Zero Retracement: Price slices right through the stop cluster, but instead of leaving a long wick, the 1-hour candle closes firmly beyond the level.
- CVD (Cumulative Volume Delta) Explodes: Aggressive market buy/sell delta rockets vertically alongside price. This signals that it's not just the exchange's automated liquidations filling orders—whales and aggressive market orders are actively joining the breakout.
- Post-Sweep Open Interest Behavior: Once the initial cascade passes, OI doesn't crash to zero (or it takes a minor dip and immediately ramps back up). This is a dead giveaway that new smart money is taking over the newly opened positions, rather than just old bad debt getting cleared.
When this plays out, look for a entry on the retest of the broken level from the outside, placing a tight stop just behind the structure.
Non-Negotiable Risk Management Rules for Liquidity Hunters
Trading liquidation sweeps is walking through a minefield. Miss your timing by two seconds or trade naked without a stop, and you'll quickly find yourself becoming someone else's heatmap data point.
- Ditch the Degenerate Leverage (>10x): When playing against volatile stop-run spikes, slippage can eat 1–2% in milliseconds. High leverage will blow up your margin before your stop loss even gets a chance to trigger on the exchange order book.
- Stops are Non-Negotiable: If price blasts through a liquidation level and keeps running without you, don't average down. Take the fixed loss (1–2% max portfolio risk), accept that the market proved you wrong, and wait for the next clean setup.
- Ignore Low-Tier Noise: Don't jump on every tiny order book blip. Whales only hunt real money. Focus exclusively on macro clusters where the liquidation volume dwarfs the asset's average 1-minute trading volume.
Trading isn't about guessing which direction the next god candle will go. It's cold-blooded probability math and taking advantage of other traders' mistakes. Analyze the liquidity map, think like a market maker, and never leave a position unprotected.
