KAIROS ROSCETTA STONE

Institutional WebSocket Data Dictionary (v7.5_mega_2665k)

Welcome to the KAIROS_DATA_DICTIONARY. This document serves as the "Rosetta Stone" for institutional clients connecting to our live execution gateway via ws://api.kairossignal.com:8090/ws.

Every signal broadcast through our WebSocket contains a zlib-compressed, base64-encoded JSON payload representing our physical-compute engine's entire state at the exact millisecond of the signal trigger. Below is the mapping for the 128D state_128d embedding array.

The Payload Schema

When you subscribe to channel:*:omni, you will receive streams matching this structure:

{
  "type": "SIGNAL",
  "asset": "BTCUSDT",
  "direction": "BUY",
  "price": 94250.50,
  "confidence": 0.825,
  "xgb_win_probability": 0.881,
  "regime": "BULL_TRENDING",
  "ev_pct": 2.45,
  "state_128d": "[BASE64_GZIP_STRING]",
  "timestamp_utc": "2026-03-10T14:30:00Z",
  "model_version": "v7.5_mega_2665k"
}

The 128D `state_128d` Vector Mapping

The state_128d array is a 128-element Float32 vector ingested directly by our XGBoost Kernel. To decode it, un-base64 the string, decompress via zlib, and map to a Float32 array. The indices represent:

[00 - 16] Core Math Layers: Tanh-normalized probability scores from the 17 physical/geometric layers (e.g., Quantum Schrödinger Wave, Kolmogorov Complexity, Betti Void Topology).
[17 - 33] Cross-Product Features: Rotational momentum dynamics between the core layers.
[34 - 49] Return Statistics: Skewness, kurtosis, and multi-window VWAP standard deviations.
[50 - 58] Liquidity & Volume: Orderbook absorption rates and Volume-Weighted Average Price (VWAP) delta.
[66 - 81] Multi-Window Alpha: Aggregated momentum over [1, 3, 5, 10, 15, 30, 60, 120] tick horizons.
[82 - 97] Score Aggregates: 75th/25th percentile spreads, entropy, and max/min bounds of the prediction ensemble.
[114 - 127] Regime Physics: GARCH(1,1) forward volatility, autocorrelation lag tables, and Hurst exponent proxies for market efficiency.

Execution Veto Logic

A signal is ONLY broadcast if it survives the 12-stage sieve. Our primary ML quality filter is the XGBoost kernel:

Confidence Gate: All signals must meet a baseline (e.g., 0.65 + volatility adjust).
XGBoost Veto: The 128D state vector is evaluated against our 5.17M trade database. If p < 0.70, the trade is killed and will never hit the WebSocket.
Causal Hard Veto: If the Do-Calculus MFDFA layer determines the market is too efficient to exploit (boost < 1.0), the signal is killed.

KAIROS Sovereign Intelligence • Institutional Operations • VER: 144.6.0-MISSION-READY