Agents 🤖FEEDS 🤖BRAIN 🤖DESK 🤖CMD API Bot 💰 Kraken Live — open LIVE
The Thesis The Numbers The Moat Roadmap Contact
btcpredictor.io / investors

The case
for building
in public.

An AI trading system, live with real money, every prediction logged on-chain before execution. Five days old. Already profitable. This is what the data looks like.

Closed Trades
0
22–27 Feb 2026
Win Rate
0
above breakeven ~51%
Net PnL
$0
on $100 capital
ROI (5 days)
0%
~1,100% annualized*
Live on Kraken Futures · On-chain verified · Open source
WR 54.4% Best bucket [0.75–0.80]: +$20.62 on 71 trades Best hour: 03:00 UTC — 87.5% WR Best day: Feb 25 — 79.2% WR, +$17.13 On-chain: Polygon PoS · 0xe4661F7...33a55 Capital at risk: $100 · No leverage on the house Architecture: Dual-gate LLM + XGBoost WR 54.4% Best bucket [0.75–0.80]: +$20.62 on 71 trades Best hour: 03:00 UTC — 87.5% WR Best day: Feb 25 — 79.2% WR, +$17.13 On-chain: Polygon PoS · 0xe4661F7...33a55 Capital at risk: $100 · No leverage on the house Architecture: Dual-gate LLM + XGBoost

BTC first.
Everything else next.

01.A
Why Bitcoin is the hardest test
BTC is the most liquid, most analyzed, most written-about asset in crypto. If a prediction system can find repeatable edge here, it can find edge anywhere. We chose the hardest market on purpose.
01.B
The inefficiency thesis
BTC is not an efficient market. It's dominated by retail emotion, leverage cascades, funding rate extremes, and narrative cycles. This creates structural micro-patterns — exploitable with the right signals.
01.C
The universal framework
The same architecture runs on any asset with order book depth + sentiment signals: ETH, SOL, mid-cap tokens (top 20–100), even FX. The real x2–x5 opportunities live in less-efficient markets. BTC proves the engine works.
01.D
Build in public as moat
The track record is on-chain. Immutable. Every prediction hashed before execution — before we know the outcome. You can't fake that. When the numbers compound, so does the credibility.
"We're not saying BTC is predictable. We're saying that in certain conditions, with certain indicators aligned, this market has structure. And we found it."

Performance data.
Unfiltered.

261 trades · 5 days · $100 capital · every row verifiable on-chain

btcpredictor_performance.log — 22 Feb → 27 Feb 2026
Metric Value Context
Overall Win Rate 54.4% Break-even ≈ 51% after fees. Above threshold.
Gold Zone WR conf 0.75–0.80 56.3% 71 trades → +$20.62 PnL. This is where we live now.
Best Single Day 79.2% WR Feb 25 · 24 trades · +$17.13 net PnL
Best Time Window 76.0% WR 03:00–09:00 UTC · Asia open + European pre-market
Best RSI Zone 61.9% WR RSI 55–70 · 84 trades · +$17.92 · 119% of total profit
Total Net PnL +$15.09 After fees ($2.07). 261 closed trades.
Audit Trail on-chain Polygon PoS keccak256 hash committed before execution. Immutable.
Architecture live Dual-gate AI Claude LLM + XGBoost must agree. Either can veto.

RECENT OPTIMIZATION: Confidence threshold raised from 0.65 → 0.75 on Feb 27. The [0.70–0.75) bucket was generating –$5.89 on 42 trades. The [0.75–0.80) bucket generated +$20.62 on 71 trades. Volume reduced ~57%, expected PnL/trade to increase ~3×.

Why this is
hard to replicate.

L1
50+ Real-Time Data Sources
Order book microstructure · funding rates · long/short ratio · CVD · taker buy/sell volume · 4 news feeds (CoinDesk, CryptoCompare, CNBC, Sole24Ore) · Fear & Greed index · multi-timeframe OHLC (5m, 15m, 4h) · Open Interest · macro event calendar. All fetched, formatted, and merged every 6 minutes.
// DATA LAYER — wf01A · 49 nodes · ~$0/month (all public APIs)
L2
Dual-Gate Veto System
Claude LLM reads the full market context. XGBoost ML scores the technical indicators. If they disagree on direction — no bet. If confidence is below 0.75 — no bet. If a macro event is detected — no bet. The system only acts when two independent intelligences align. That's the dual veto.
// DECISION LAYER — wf01B · LLM + XGB 86% accuracy on technicals
L3
Pattern Memory & Self-Calibration
The bot reads its own recent decisions before each new signal. Anti-noise filter prevents betting on the same micro-pattern twice. Rolling win rate monitoring triggers automatic bet-size reduction if performance degrades. The system adapts.
// INTELLIGENCE LAYER — pattern memory · rolling WR circuit breaker · error context
L4
On-Chain Immutable Audit Trail
Every prediction is hashed (keccak256) and committed to Polygon PoS before execution. The contract timestamp is before the trade. You can independently verify that every signal in our track record was real — not cherry-picked post-hoc. This is what separates us from every other "backtested" system.
// AUDIT LAYER — BTCBotAudit.sol · Polygon PoS · <$0.001/tx
"The code is open source. The map is free. The experience of walking the road is not."

Four phases.
One direction.

▶ Now
01
Beta — Proof of Edge
$100 live deployment. Real trades. Real losses possible. Dual-gate AI running every 6 minutes on Kraken Futures. Everything on-chain.
Live since Feb 22, 2026
Next
02
Scale — MEXC + Multi-Asset
$10K+ deployment on MEXC (near-zero fees, MX token discount, 3% APR on collateral). Add ETH, SOL, mid-cap tokens. Optimize OI + volume signals.
3–6 months
Building
03
SaaS — Multi-User Signals
Each user connects their own API keys. Bot executes on their behalf. Real-time signals. Subscription model: Free (delayed) / Pro (live) / Enterprise (API).
6–12 months
Vision
04
Alpha Marketplace
On-chain track record as credibility layer. Community voting on signals. Institutional API access. Verified performance becomes the product.
12+ months

This is not a pitch deck.
It's an invitation.

We're not raising a traditional round. We're building in public and inviting people who get it to be part of it.

Direct email: [email protected]  ·  Or use the form below.
Send a message directly
What we're not saying.
On-chain contract — Polygon PoS 0xe4661F7dB62644951Eb1F9Fd23DB90e647833a55

Verify independently: search the contract on PolygonScan. Every commitment timestamp precedes its trade timestamp. That's the claim. The blockchain is the proof.