⚔️ BTC BOT // WAR ROOM

00:00:00

UTC

BTCPREDICTOR.IO · OPEN SOURCE · AI + BLOCKCHAIN

AUTONOMOUS BTC PREDICTION ENGINE

We don't predict the market. We read its behavior.

Claude Sonnet + XGBoost · Dual-Gate AI · Every prediction committed on-chain BEFORE execution

—

BTC/USD · LIVE

LAST UPDATE: —

POLYGON PoS · VERIFIED ✓ ON-CHAIN AUDIT TRAIL

LIVE DATA

BOT ALIVE

Checking...

Where the edge lives

85.2%

Win Rate in the Golden Window

UP direction · confidence 0.65–0.70 · US hours 8/13/14/15/21 UTC. 54 samples / 60 days. +21% cumulative.
Overall ghost WR 47% — we don't pretend otherwise. Our Council fires only inside the window.

Beta — invite only

AI FEEDS · 16 SOURCES

KRAKENPRICE

BINANCEKLINES

COINGECKOMARKET

FUNDING RATEDERIVATIVES

L/S RATIODERIVATIVES

FEAR & GREEDSENTIMENT

CRYPTOCMPNEWS

COINDESKNEWS

OPEN INT.DERIVATIVES

TAKER FLOWFLOW

COINTELEGR.NEWS

XGBOOSTML GATE

PATTERNSMEMORY

CNBCNEWS

ORDER BOOKMICROSTRUCTURE

MULTI-TFTECHNICAL

INFO

PERIOD

Railway API non raggiungibile — dati non aggiornati ✕

⚠️ — ✕

—

claude-sonnet-4-6 · weight 0.30

No deliberation data yet.

—

gemini-2.0-flash · weight 0.15

No deliberation data yet.

—

xgboost-local · weight 0.25

No deliberation data yet.

Phase State Ising Model

DISORDER

FR 0.00

LS 0.00

OI 0.00

Market agents are scattered — no directional consensus. The bot observes.

WR%

WIN RATE

—

P&L

—

F/G

FEAR/GREED

—

#BET

BETS

—

DRAWDOWN

—

STK

STREAK

—

BOT

—

checking...

TODAY PnL

—

bets today

LAST SIGNAL

— conf — — —

SIGNAL LOG BOARD

SIGNALS

—

BETS

—

WIN RATE

—

AVG CONF

—

AVG PNL/BET

—

#	DATE/TIME	PREDICTION	CONFIDENCE	ENTRY $	EXIT $	RSI	EMA	SCORE	SIZE	TYPE	RESULT	PNL $ / THR%	FEE $	SL/TP	⛓

BTC Price

—

Last entry

Total Signals

—

All time

Accuracy

—

Correct pred.

Total PnL

—

from 2026-02-24

APR annualized

—

— days active

Session ROI

—

since session start

Win Rate

—

Executed trades

Signal WR

—

All predictions verified

Fear & Greed

—

Bets Taken

—

Out of total

Max Drawdown

—

Single trade

Expectancy

—

avg $/bet

Profit Factor

—

wins/losses

Cumulative Equity Curve (Real Bets)

Avg Confidence ⓘ

—

Best Trade PnL ⓘ

—

Current Streak ⓘ

—

Avg Entry Slippage ⓘ

—

Avg RR Ratio ⓘ

—

Total Fees ⓘ

—

Total Slippage ⓘ

—

APR Today ⓘ

—

APR 7d ⓘ

—

Best Streak

—

Hit Rate 7d

—

Conf: WIN/LOSS

—

BTC/USDT Live Chart

▲ WIN ▼ LOSS ◆ OPEN

CHART UNAVAILABLE

Rolling Win Rate (BETs)

— WR-10 — WR-20 — 50%

oldest betnewest bet

Rolling Win Rate (20-Bet Window)

Confidence Score — Every Bet

Prediction Galaxy ✦

CONF × PnL

← low confconfidencehigh conf →

Signal Quality

LIVE

Confidence Distribution

BETs ONLY

Confidence Over Time

ALL

oldestnewest

PnL per Bet ($)

REAL BETS

oldestnewest

Auto-Training — XGBoost Engine

Dual-gate AI: XGBoost learns from live bet history. Retrains automatically every Sunday 03:00 UTC. Both XGB + Claude must agree before placing an order.

New Bets Since Retrain

—

/ 30 target

0% toward next meaningful retrain

Direction Accuracy

—

— signals in dataset

Last Retrain

—

Next Scheduled

—

Sunday 3:00 UTC · auto

Trading Stats — Expectancy

Expectancy = avg profit per bet. >$0 = edge positive. L20 = last 20 bets trend. WR split reveals UP vs DOWN directional bias.

Expectancy

—

per bet (all time)

Expectancy L20

—

last 20 bets

Avg Win / Avg Loss

—

reward/risk ratio

UP Bets WR

—

PnL: —

DOWN Bets WR

—

PnL: —

Avg PnL / Bet

—

wins: — / losses: —

Win Rate per UTC Hour

■ >55% WR ■ <45% WR ■ neutral / no data click cell for detail · UTC hours 00–23

🤖

Automation Agents

17 n8n workflows orchestrate the entire trading pipeline — from data ingestion to AI prediction, trade execution, Telegram notifications, on-chain audit, and social media publishing. The launchd agents run locally on Mac for maintenance, reporting, and orphan bet rescue.

n8n Workflows

Premi Refresh per caricare i workflow n8n.

launchd Agents — Mac Local

Scheduled Tasks

LOCAL · macOS launchd

AGENT	LABEL	SCHEDULE	NEXT RUN	LOG	STATUS

🔬

Walk-Forward Backtest

Out-of-sample simulation across 6 confidence strategies (A→F). Each strategy is tested on unseen data to identify the optimal threshold — the one that maximizes PnL without overfitting to historical signals.

Ultimo backtest: caricamento...

Best Strategy

—

Current (FULL_STACK)

—

Test WR

—

Profit Factor

—

Best strategy

XGB CV Acc

—

train set

Max Streak

—

W / L

Equity Curves — Test Set

per strategia, partendo da $0

Strategy Comparison

WALK-FORWARD

Strategia	N	WR%	PnL	$/bet	MaxDD	PF	Sortino	Sharpe

Confidence Calibration

CALIBRATION

Bucket	N	ExpWR	ActWR	Δ	$/bet

24h Heatmap — Win Rate

BACKTEST

■ HOT ≥60%■ DEAD <45%■ no data

▸ Walk-Forward Report (testo grezzo)

Loading...

▸ XGBoost Model Report

Loading...

💸

Operational Costs

Real-time breakdown of all costs to run the bot: Kraken trading fees, AI APIs, n8n VPS, Railway hosting, and Claude Code. Build cost: $709.48 (Feb 2026, Claude Pro). From 1 Mar 2026: Claude Max plan (€218.11/mo, unlimited sessions, Opus 4.6). Infra: ~$11/mo. Profit counts only after all costs are covered.

Total Cost MTD

—

Kraken Fees

—

n8n Plan

—

Claude API

—

Claude Code

—

Railway

—

RAILWAY BUDGET — Hobby Plan aggiorna manualmente

$4.87

CREDITS LEFT

$0.04

BURN/DAY

$1.30

PROJ/MONTH

$0.13 used (2.6%) $5.00 credit/mese

SUBSCRIPTION EXPIRY

Cost Breakdown

🛠 Development History — Build Cost

ONE-TIME

Total Build Cost

$709.48

Claude Code CLI sessions

Lines of Code

29,683

added · 4,755 removed

Git Commits

409

main branch

Build Duration

~30d

Feb 2026 · from scratch

Session Breakdown

Session	Model	Cost	API Time	Code Changes	Note
94969f8e	claude-sonnet-4-6	$709.48	1d 4h 29m	+29,683 / -4,755	Full system build · Feb 2026
1 Mar 2026+	Claude Max	€218.11/mo	illimitato	20× usage vs Pro	Opus 4.6 · sessioni illimitate
TOTAL BUILD COST		$709.48	one-time · ongoing: €218.11/mo Claude Max + ~$11/mo infra

💡 Build in Public. This system was built entirely with Claude Code CLI in February 2026, starting from a whiteboard sketch. The $709.48 is the full transparent cost of development — no hidden expenses. Monthly operational cost going forward is ~$11/mo (VPS + Railway + domain).

🙏 Support the Build

30 days. 409 commits. $709.48 in API calls. Built from a whiteboard sketch to a live trading system — line by line, dollar by dollar, all in public.
If you find value in the radical transparency, the on-chain audit trail, or simply want to keep the server lights on — a crypto donation goes directly into operational costs. Every MATIC received is another prediction committed immutably to Polygon. ∎

Accepts: MATIC · ETH · USDC · USDT · Network: Polygon PoS

Polygon Wallet

0x7Ac896…0d51f0

View on Polygonscan ↗

📈

Edge Analysis — Expectancy

Expectancy measures how much the bot earns on average per bet. A positive value confirms a real statistical edge. Rolling L20 shows whether that edge is stable or degrading over the most recent 20 trades.

Expectancy / Trade

—

E = WR x avgWin - (1-WR) x avgLoss

Profit Factor

—

gross wins / gross losses

Kelly %

—

optimal fraction

Win Rate

—

executed bets

Expectancy — Formula Breakdown

Avg Win

—

per winning bet

Avg Loss

—

per losing bet

Win/Loss Ratio

—

avgWin / avgLoss

Total Bets (closed)

—

sample size

Gross Wins

—

sum of profitable trades

Gross Losses

—

sum of losing trades

▲ UP Bets

—

Win Rate —

Total PnL —

Count —

▼ DOWN Bets

—

Win Rate —

Total PnL —

Count —

Rolling Expectancy — Last 20 Bets

—

Expectancy over the last 20 closed bets.
Positive = system profitable in the short term.

🧠

XGBoost Model — Training & Status

The dual-gate prediction engine trains a gradient-boosted classifier on historical signals. Direction model targets UP/DOWN accuracy; correctness model predicts whether the LLM will be right. Both must agree before a bet is placed.

Auto-Training — XGBoost Engine

Bets toward next retrain —

— / 30 new bets

🎯

CV Accuracy ⓘ

—

— signals

📅

Last Retrain ⓘ

—

⏱

Next Scheduled ⓘ

—

Sunday 3:00 UTC

🔐

Confidence Thr. ⓘ

—

min to open bet

Model Status

LIVE

XGB Direction —

XGB Correctness —

Base Position Size —

Dead Hours UTC

—

Automated Pipeline

LAUNCHD · DOM 3:00 UTC

1.build_dataset.py — refresh dataset from Supabase
2.train_xgboost.py — retrain XGBoost
3.backtest.py — walk-forward validation
4.analyze_errors.py — error pattern analysis
5.git push → Railway auto-deploy
6./reload-calibration → hot reload

▸ XGBoost Model Report

Loading...

▸ Walk-Forward Backtest Report

Loading...