Analytics Streams
The platform runs 14 real-time analytics feeds, produced by services/analytics_worker.py and backed by Redis for sub-second read latency. These feeds are consumed both by the strategy layer (as model features) and directly exposed through the MCP introspection layer for observability.
| Stream | What it measures |
|---|---|
| Regime | Current market state (trending / mean-reverting / volatile / quiet) via the HMM model |
| VPIN | Volume-Synchronized Probability of Informed Trading — order-flow toxicity |
| OI Velocity | Rate of change in open interest |
| CVD | Cumulative Volume Delta — buy/sell pressure |
| Volume Profile | Point-of-control and volume distribution by price level |
| Breadth | Market-wide participation across tracked symbols |
| Factors | Factor decomposition of price movement |
| Session Intelligence | Time-of-day / session-relative context |
| Correlation | Cross-asset correlation tracking |
| Cross-Exchange | Spread and lead-lag tracking across venues |
| Fractal | Self-similarity / multi-timeframe structure |
| Indicators | Standard technical indicators (RSI, MACD, ADX) computed on the platform's own pipeline |
| Trade Readiness | Composite score combining the above into an entry-quality signal |
| Feeder Health | Staleness and uptime tracking for every upstream market-data feed |
Why this matters
Every strategy decision is made against live, multi-dimensional market context — not price alone. Feeder Health specifically exists so that a stale or disconnected upstream feed degrades the system's confidence in a signal rather than silently feeding bad data into a trade decision.