Antigravity AI — Live Crypto Gainer ML Predictions

📡 Active Predictions 0

Coin	Confidence	Score	Entry	Take Profit	Stop Loss	Unrealized	Signals	Picked (EST)
Loading live data...

⚡ Alpha Engine Picks 0 from Alpha Engine

Symbol	Direction	Strategy	Entry	TP	SL	P/L %	Grade	Entry Date
Loading Alpha Engine data...

🔥 Crypto Gainer ML 0 from Crypto Gainer ML

Coin	Score	Entry	TP	SL	P/L %	Picked (EST)
No crypto gainer signals yet...

📋 Resolved Trade Log 0

Coin	Outcome	Entry	Exit	P/L %	Score	Picked (EST)	Resolved (EST)
Picks appear here as they resolve...

🔬 Loss Forensics — Why 24 Picks Lost Deep Analysis

🔎 Root Cause Analysis: 34 Picks, 23.5% Win Rate, -28.49% P/L

Key Finding: SELL signals on 15m were 85.7% accurate (6/7 TP hits). BUY signals on 1h were 8.3% accurate (1/12). The model predicts drops better than rallies in bearish conditions.

📍 By Direction

Direction	Picks	Wins	WR	Avg P/L
BUY	27	1	3.7%	-1.14%
SELL	7	6	85.7%	+0.73%

⏰ By Timeframe

Timeframe	Picks	Wins	WR	Avg P/L
15m	22	7	31.8%	-0.46%
1h	12	1	8.3%	-1.78%

✅ v1.4 Model Tweaks Applied Based on This Analysis

Fix	Before	After	Impact
Confidence gate	0.60	0.65 (0.70 for 1h)	Blocks 19 coin-flip picks
Model selection	max prob (overfit)	C_random_forest	Stops rewarding overconfident models
BTC regime	4h candle only	4h OR 12h OR EMA	Catches slow BTC bleeds
Per-symbol dedup	unlimited	max 1 per coin	Prevents BNB x3 / ZRO x2 disasters
TF conflict check	none	block opposite signals	Stops LINK SELL+BUY contradictions
Trend alignment	none	EMA20 vs EMA50	BUY must agree with trend (or 0.75+ conf)
SL distance	0.5% min	0.8% min	Prevents ZRO-style -6.73% gap-through
TP/SL ratios (scalp)	3.0/1.5 ATR	2.5/2.0 ATR	More achievable TP, fewer noise SL exits

Estimated impact: Of the original 34 picks, only 9 would pass v1.4 gates. Projected win rate: 45-55% (vs 23.5% actual).

📊 Backtest — Real Binance Kline Data Loading...

Loading real backtest results from walk-forward validation on actual Binance kline data...

🔍 Transparency Report — Honest Assessment

⚠️ What You Need To Know Before Following These Picks

✅ What IS Proven (Pattern Discovery)

We analyzed 182,500 data points across 100 coins over ~5 years and found 30 statistically significant pre-pump precursor conditions:

Williams %R Oversold (significance: 245.7) — price near bottom of range before explosion
Doji Candle Formation (significance: 238.3) — indecision = coiled spring
Low Range Position (significance: 226.8) — price compressed near lows
Elevated MFI (significance: 211.8) — smart money flowing in while price flat
TTM Squeeze Active (significance: 144.6) — BB inside KC, volatility compression
EMA 9/21 Cross (significance: 157.7) — short-term trend shifting bullish

These are well-known institutional patterns. The discovery is legitimate.

❌ What IS NOT Proven (Backtest)

The real walk-forward backtest on Binance klines shows aggregate Sharpe of -10.44 and 35.4% win rate across 4,031 trades. The models are not yet profitable in backtesting. Training data quality needs improvement — see real backtest results.

📊 ML Model Metrics

AUC-ROC: 0.667 — above random (0.50) but below production-grade (0.75+)
F1 Score: 0.319 — low, high false positive rate
Training: 182,500 synthetic samples, NOT real candles

💰 Should You Trade With Real Money?

NO. Absolutely not.

System live since Feb 23, 2026 (~2 weeks)
32 resolved picks (18 TP / 14 SL)
Current live win rate: 56.2%
Need 50+ resolved picks for statistical significance

Recommendation: Paper trade for 2–4 weeks. If win rate > 40% across 50+ picks with positive P/L, then consider small positions.

🧠 ML Strategy Architecture

📊 Volume Accumulation

30%

RVOL > 2x average with small bodies = smart money loading.

🔒 Volatility Compression

25%

ATR below 80% of 50d avg, BB squeeze. Tight ranges precede breakouts.

🚀 Momentum Building

20%

MACD histogram rising, RSI 40-65 sweet spot.

💥 Breakout Proximity

15%

Price within 2% of 20d high with bullish EMA alignment.

📈 OBV Divergence

10%

OBV accumulating while price flat — institutional accumulation.

⚙️ Risk Management

2.5:1

TP = 2.5x ATR proxy, SL = 1x ATR. 48h expiry. Max 10 concurrent.

🤖 Model Architecture

Component	Detail
Ensemble	XGBoost + LightGBM + Random Forest + Logistic Regression
Features	139 features: momentum, volume, volatility, price structure, trend, temporal
Training	Real Binance klines, 36 pair/TF combos retrained with v1.5 research overhaul
v1.5 Enhancements	Isotonic calibration • Purged walk-forward CV (75/25 + 20-bar gap) • Early stopping (50 rounds) • Meta-labeling filter (M2 trade/no-trade) • SMOTE disabled • Reduced complexity (300 trees, depth 4-5)
Backtest	Walk-forward on real Binance klines — see results below
Automation	GitHub Actions every 4h → CoinGecko → score → pick → TP/SL → Discord
Discord	GOOGLE GEMINI — REVERSE ENGINEERED DAILY TOP GAINERS STRAT

Antigravity ML Predictions