1. Maintain trustworthy market observations
The gateway connects to Binance Spot BTCUSDT aggregate trades and Level 2 depth updates. It builds a synchronized local order book from a REST snapshot plus sequenced updates. Provider IDs, timestamps, products, ranges, and decimal values are checked before use.
The UI never parses exchange payloads. It consumes a stable MarketFrame contract so presentation cannot silently redefine the market mechanics.
2. Normalize flow and depth
Trade notional is grouped by taker direction over a rolling 15-second window. Nearby bid and ask notional is measured within 20 basis points of the midpoint. Each component is normalized so direction and relative imbalance can be combined on a common scale.
Model mp-v1.1.0 uses 70% trade-flow pressure and 30% book pressure. These parameters belong to the versioned model, not the animation.
3. Stabilize without predicting
Exponential smoothing reduces frame-to-frame noise. Hysteresis prevents the label from flipping repeatedly around a threshold. An activity measure helps distinguish energetic flow from a quiet, fragile reading.
These operations make the display readable; they do not estimate future returns or introduce an AI price forecast.
4. Map the verified frame to the tug of war
Positive pressure means Buyers and maps left; negative pressure means Sellers and maps right. The invariant is visualX = -pressure / 100 × maxVisualOffset. A zero value stays centred.
If the frame is stale or fixtures are configured in production, the product fails closed. It shows a delayed or unavailable state and pauses directional movement instead of displaying unverified continuity.