About Kairos Signal

ESTABLISHED 2025

Kairos Signal is a quantitative signal engine. It generates statistical signals for 47 crypto perpetual markets using a 63-layer Neural SDE called ByteDAG. The entire system runs on self-owned infrastructure. No cloud, no VC.

What It Does

ByteDAG runs a 256-to-256 forward pass with a 512-wide whiteboard feature space. The vector the API actually returns is 32 values. The model ingests raw market tick data stored in ClickHouse and produces directional signals for 47 crypto perpetual futures contracts on Hyperliquid.

No historical performance or return figure is claimed; prior backtest figures were withdrawn as non-reproducible.

The architecture is informed by peer-reviewed mathematical sources spanning stochastic calculus, gauge theory, and topological data analysis. The layers are implemented in PyTorch, trained on real data, and tested empirically.

The 63-Layer Architecture

All raw tick data feeds into the ByteDAG model. The architecture is split into three main phases:

  • Neural SDE Flow Integration (Layers 1-29): Volume-preserving continuous neural mappings that track temporal state anomalies.
  • Fractal & Stability Classification (Layers 30-33): Lyapunov exponents and subdifferential boundaries determining regime shifts.
  • Wilson Loop Gate (Layers 34-37): Obstruction filters that score and output high-conviction signals.

The remaining layers handle residual connections, normalization, and output projection.

By the Numbers

  • 63 layers in the ByteDAG neural SDE
  • 256-to-256 internal forward pass (the published vector is 32 values)
  • 512-wide internal whiteboard feature space
  • 47 crypto perpetual futures covered
  • Raw market tick data in ClickHouse
  • 1,902 blog posts published
  • Peer-reviewed mathematical sources referenced
    • (no performance figure claimed)

What It Is Not

This is not investment advice. Kairos Signal is not a broker-dealer, investment advisor, or financial services provider. It is a machine learning research product. All outputs are for research and informational purposes only.