Agents πŸ€–FEEDS πŸ€–BRAIN πŸ€–DESK πŸ€–CMD API Bot β€” πŸ’° β€” Kraken Live β€” β€” open LIVE

btcpredictor.io — AI Architecture

THE DUAL-GATE AI

How Claude LLM + XGBoost decide every Bitcoin trade

1,070+ signals analyzed 54% win rate 86% direction accuracy
Explore the system
Scroll

The Problem

"Financial markets generate signals constantly. Most of it is noise."

"An AI model that can't tell signal from noise loses money β€” even if it's 'intelligent'."

"The solution: two independent gates. Both must open. Only then is the trade executed."

Gate 1: Claude LLM

LIVE on every hourly signal
CLAUDE
GATE 1
conf ≥ 0.75 → PASS
conf < 0.75 → SKIP

Claude analyzes 13 data sources in natural language β€” not just numbers, but context, narratives, and cross-asset correlations.

Output: a direction (UP or DOWN) and a confidence score between 0.0 and 1.0. Only signals with confidence ≥ 0.75 reach the second gate.

The confidence score isn't arbitrary: it scales into a position sizing multiplier. The higher it is, the larger the bet β€” within a clamped range of 0.8x to 1.2x base capital.

// 13 INPUT SOURCES β€” analyzed by Claude every cycle

01  FEAR_GREED_INDEX      β†’ sentiment 0-100
02  RSI_14                β†’ momentum indicator
03  TECHNICAL_BIAS        β†’ bullish / bearish score
04  MACRO_NEWS            β†’ geopolitical events
05  VOLUME_ANALYSIS       β†’ market depth
06  MARKET_STRUCTURE      β†’ support / resistance
07  FUNDING_RATE          β†’ perpetual futures signal
08  OPEN_INTEREST         β†’ leveraged positions
09  BTC_DOMINANCE         β†’ BTC market cap share
10  ON_CHAIN_FLOWS        β†’ exchange inflow / outflow
11  LIQUIDATION_MAP       β†’ leverage imbalance
12  STABLECOIN_FLOWS      β†’ smart money movement
13  CORRELATION_INDEX     β†’ cross-asset momentum

XGBoost β€” The Verifier

XGBOOST ROOT n_estimators: 200 conf > threshold? fear_greed < 50? rsi14 > 60? LEAF: LONG LEAF: SHORT LEAF: LONG LEAF: SHORT YES NO

XGBoost (eXtreme Gradient Boosting) is a machine learning algorithm that learns from historical patterns β€” not hand-written rules.

The model was trained on 1,000+ past signals with their actual correctness as training labels. Every new Claude prediction is evaluated against patterns that have historically worked.

Validation uses TimeSeriesSplit β€” not a random train/test split, because the future can never inform the past.

// MODEL CONFIGURATION

Model           XGBClassifier
n_estimators    200
max_depth       4
learning_rate   0.05
Features        11  (temporal + technical)
Validation      TimeSeriesSplit(n_splits=5)
Retrain         on demand (every ~500 signals)

// Features used

confidence fear_greed_value rsi14 technical_score hour_sin hour_cos technical_bias_bullish signal_fg_fear dow_sin dow_cos session

hour_sin/cos and dow_sin/cos encode temporal cyclicality without discontinuities.
session: Asia=0, London=1, New York=2.

Gate 2: The XGBoost Vote

LLM dir = XGB dir → TRADE
LLM dir ≠ XGB dir → SKIP
GATE 2
XGBOOST

XGBoost votes: does Claude's predicted direction match historically successful patterns?

If yes → the trade is executed on Kraken Futures (PF_XBTUSD).

If no → the signal is silently discarded (SKIP). No trade, no risk.

This high selectivity β€” about 7% of signals pass both gates β€” is intentional. It's better to do nothing than to do the wrong thing.

0+ signals analyzed
~7% pass both gates
0% direction accuracy

On-Chain Audit

Every signal that passes both gates is permanently written to the Polygon PoS blockchain before the trade is executed.

Immutable. Verifiable. Tamper-proof.

This is the ultimate audit trail: nobody can claim retroactive predictions or alter signal history. The on-chain timestamp always precedes trade execution.

Smart Contract 0xe4661F7dB62644951Eb1F9Fd23DB90e647833a55 Polygon PoS Mainnet

The Dual-Gate Flow

From data source to trade β€” every step, in order.

BTC MARKET (Kraken Futures)
DATA COLLECTION β€” 13 sources in parallel
CLAUDE LLM β€” Analysis & Confidence Score
YES · conf ≥ 0.75
NO
XGBOOST VERIFICATION
YES · they agree
NO
TRADE EXECUTED
ON-CHAIN COMMIT (Polygon)

Watch the system in action

Live dashboard β€” real-time signals, full history, metrics.

// Legal Disclaimer β€” Read Before Proceeding

This site does not provide financial advice. Predictions are experimental and not guaranteed. Past results are not indicative of future returns.

Trading derivative instruments (futures, perpetual contracts) carries significant risk of capital loss, including total loss of investment. The use of leverage amplifies both gains and losses.

btcpredictor.io is an experimental research and development project. It is not registered as a financial advisor, fund manager, or signal provider under MiFID II or any other applicable financial regulation. Any investment decision is the sole responsibility of the user.

Updated: February 2026 — Version 1.0