🔍 Data Truth
Audit data sources · quarantine bad training data · wire real prices · --
Current data mode
--
Findings (mock)
--
Findings (live)
--
Training rows (mock)
--
🛠 The fix: 4 steps
1

Quarantine contaminated training data

Every existing training row was generated from synthetic mock prices. The brain has been training on fake patterns. Quarantine them (move to a separate keyspace) so the brain forgets. Your model will reset to default weights.

2

Wire a real price feed

The site supports Finnhub (free tier, real-time US stocks), Polygon, Tradier, and Alpaca. Open Settings, pick a provider, paste your API key. Finnhub free is the easiest start — 60 calls/min and real US trades.

⚙ Open Settings 📖 Step-by-step guide
3

Verify the demo banner disappears

Once a real feed is connected, the red DEMO MODE banner at the top of every page will turn off automatically. window.BPLEONE_DATA_MODE flips from 'mock' to 'live'. From that point forward, every brain finding is tagged dataSource: 'live' and is eligible for training.

4

Re-seed the brain with real outcomes

Resolve some trades with real prices in Journal PRO or let Autopilot run for a few days against the live feed. After ~20 real labeled trades, the brain has a fresh, honest training set.

🎯 Reality check — override mock prices manually
Paste current prices for the symbols you care about. The mock engine will use these as seeds and stop drifting. Useful for testing without a real feed.
SYMBOL
MOCK PRICE
YOUR PRICE
DELTA
NOTE
💀 Danger zone
Nuclear options. Use when the brain is so corrupted you want a clean slate.
📋 Why this matters
The whole point of a self-learning brain is to find real patterns. If it trains on simulated random-walk prices, it learns the structure of the simulator — not the structure of the actual market. Predictions will look confident but be meaningless.

Symptoms you've been training on bad data:
• Brain confidence is high (70%+) but real outcomes are random.
• Per-symbol weights drift to extremes — every ticker becomes "1.5x" or "0.5x".
• Feature weights have flipped sign multiple times.
• Hit rate on real trades doesn't match what the brain claimed.

The fix is to start fresh with real data. The Data Truth audit is your way to do that without losing all the platform infrastructure.