// section 01
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."
// section 02 β gate 01
Gate 1: Claude LLM
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
// section 03 β the verifier
XGBoost β The Verifier
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
hour_sin/cos and dow_sin/cos encode temporal cyclicality without discontinuities.
session: Asia=0, London=1, New York=2.
// section 04 β gate 02
Gate 2: The XGBoost Vote
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.
// section 05 β immutability
On-Chain Audit
PASS
GATE
SIGN
PoS
CHAIN
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.
// section 06 β full flow
The Dual-Gate Flow
From data source to trade β every step, in order.
DISCARDED
Watch the system in action
Live dashboard β real-time signals, full history, metrics.
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