THIS IS NOT INVESTMENT ADVICE. Kairos Signal is not a broker-dealer, investment advisor, or financial services provider. This is a machine learning research product — a geometric expression of a rough approximation of the market’s hyper-dimensional phase space, from which α (statistical excess signal) can be mapped and extracted. “Alpha” refers exclusively to the mathematical α coefficient in regression residuals, not guaranteed returns. Built by independent ML researchers. All outputs are for research and informational purposes only.

63-Layer Symplectic Neural SDE · ByteDAG

Statistical Signals for Crypto Markets

A 63-layer Symplectic Neural SDE called ByteDAG generates statistical signals for 47 crypto perpetual markets . Every signal is hashed to a SHA-256 chain for independent verification.

4.3B Market Ticks
63 DAG Layers
3072D Manifold State
47 Assets Tracked
Request Credentials
Download Enterprise Deck (PDF) Technical Manual

All signals, predictions, and weights are logged in ClickHouse with SHA-256 chain hashes.

Data Sources: NOAA Weather/Storm EIA Energy Grid County Assessor Deeds DePIN Blockchain Telemetry Municipal Court Filings
Verify the Ledger

Signal Proof Feed

Every signal is hashed into a continuous SHA-256 chain in ClickHouse. 154 ledger entries and counting. Below are recent validated signals from the V103 signal validator.

Signal Validation Results Signal Validation: $1,000 → $1,863.04 (+86.3%)
Asset Direction Validation Performance Cumulative
SOL/USD LONG Target reached +$275.20 $1,112.67
XRP/USD LONG Time horizon +$234.09 $1,646.14
ETH/USD LONG Time horizon +$216.90 $1,863.04
BTC/USD LONG Time horizon +$183.48 $1,412.04
DOGE/USD LONG Time horizon +$115.90 $1,228.57
System Health

Active Network Telemetry

Real-time ingestion metrics from the data pipeline. 4.34 billion market ticks in ClickHouse.

27 Stripe SKUs Live
⚡ 1,902 Blog Posts Published
🗄️ 154 SHA-256 Ledger Entries
49K Cities Covered
Pipeline

How Data Becomes Signals

Raw market data, public records, and blockchain telemetry are processed through the 63-layer ByteDAG to produce statistical signals.

Step 1

Data Ingestion

Continuous collection of market ticks, weather data, energy grid status, county assessor filings, and DePIN blockchain telemetry.

→
Step 2

Feature Extraction

Raw data is normalized and embedded into 512-dimensional whiteboard features. Property records, lead scores, and market state vectors are computed.

→
Step 3

63-Layer Forward Pass

The ByteDAG Neural SDE runs a 3072-dimensional forward pass through 63 layers, computing symplectic dynamics, gauge field geometry, and regime classification.

→
Step 4

Signal Output + Hash

Classified signals are written to ClickHouse and hashed to a SHA-256 chain. Every signal is independently verifiable.

Data Sources

Data Sources

Multiple public and licensed data feeds cross-referenced through the 63-layer ByteDAG.

EIA Energy Grid

Tracking high-voltage utility status and local substation data from the US Energy Information Administration.

NOAA Weather & Storm

Real-time storm path modeling, wind velocity monitoring, and hail envelope extraction from NOAA feeds.

County Assessor Deeds

Tracking property transfers, liens, mortgages, and assessor valuation histories. 975K property records, 4.85M leads, 767K lead scores.

Municipal Court Filings

Direct collection of tax defaults, probate decrees, foreclosure dockets, and HOA actions from municipal courts.

Security

Security

How signal integrity is ensured through cryptographic hashing and transparent logging.

Infrastructure
Access: Key-based authentication + 2FA Audit: All signal state changes logged to ClickHouse Infrastructure: Self-hosted on owned hardware

Every signal, prediction, and hash is logged to an immutable ClickHouse ledger for full auditability. Infrastructure is self-hosted with transparent logging.

Signal Integrity
SHA-256 Chain: H(i) = Hash( H(i-1) || State(i) ) Storage: ClickHouse columnar database, 4.34B+ rows Ledger: 154 signal entries and counting

Every classified signal is hashed and chained, generating a verifiable history of signal state changes over time. The hash chain is the proof.

Products

Three Products, One Engine

The ByteDAG Neural SDE generates statistical signals, classifies property leads, and powers three API tiers from a single forward pass.

Statistical Signal API

Access the 63-layer ByteDAG forward pass output for 47 crypto perpetual markets. Three tiers: Free, Pro ($299/mo), and Quant ($499/mo).

Property Lead Data

Pre-vetted distress, foreclosure, and property leads classified into 4 quality tiers. 975K property records, 4.85M leads, 767K scored leads across 49,168 cities and 3,235 counties.

NATIONWIDE_COUNTY_ATLAS

Coverage data for 49,168 cities, 59 states/territories, and 3,235 counties. Cross-referenced property, court, and assessor data in a single API.

Start Building →
Applications

What You Can Build

Three areas where the ByteDAG signal pipeline provides useful data.

Distressed RE

Crypto Market Signals

The ByteDAG Neural SDE generates directional signals for 47 crypto perpetual markets. Signal validation: $1,000 baseline, $1,863.04 cumulative (+86.3%), 10 validated signals, 5 currently active.
  • 47 assets
  • 63-layer symplectic forward pass
  • SHA-256 chained signal proof
DePIN

DePIN Data Analysis

Public blockchain telemetry from Helium, DIMO, and Hivemapper (public data, not operated by us) cross-referenced with other data sources for infrastructure analysis.

  • Helium, DIMO, Hivemapper public data
  • Node telemetry correlation analysis
  • Geographic coverage mapping
Insurance

Property & Lead Intelligence

County assessor deeds, municipal court filings, and NOAA weather data cross-referenced to identify property distress signals. 975K property records, 4.85M leads, 767K scored leads.

  • 49,168 cities, 3,235 counties
  • 4-tier lead classification
  • Tax lien + probate + code violation scoring
Get Started — Free Tier Available →
API Example

Sample API Response

Every API response includes signal metadata and a SHA-256 chain hash. Sample response from the ByteDAG V103 signal validator:

GET /api/v2/signal/latest
{
  "signal_id": "sig_2026_06_27_1200_az",
  "timestamp": "2026-06-27T12:00:04Z",
  "dag_layers_traversed": 63,
  "state_dim": 3072,
  "classification": "TREND_BULL",
  "confidence": 0.847,
  "asset": "BTC/USD",
  "direction": "LONG",
  "sha256_chain": "a1b2c3d4...",
  "provenance": {
    "bibliography_sources": 47,
    "dag_layers": 63,
    "manifold_dim": 3072,
  }
}

What You Get

SHA-256 Chain

Every signal links to the previous via SHA-256. Tamper-evident by construction. 154 ledger entries.

Full 63-Layer Traversal

All 63 layers fire on every request. 3072-dimensional forward pass. No shortcuts.

Provenance Metadata

Every response carries signal metadata: 47 peer-reviewed sources, 63 DAG layers, 3072D manifold state.

Three API Tiers

Free, Pro ($299/mo), and Quant ($499/mo). All tiers access the same 63-layer engine.

Infrastructure

The Stack

Each component serves a specific purpose in the pipeline.

Tech Stack: ClickHouse, XGBoost, PyTorch, Go, Python, NumPy, SciPy

ClickHouse

Columnar database for high-throughput ingestion. 4.34 billion market ticks. Every signal, prediction, and hash — queryable in milliseconds.

XGBoost Veto Gate

Gradient-boosted classifier trained on millions of market state embeddings. Acts as the final signal quality gate before emission.

PyTorch + NumPy + SciPy

The 63-layer ByteDAG runs in PyTorch with NumPy/SciPy for matrix exponentials. Symplectic integrators preserve phase-space volume.

Go Ingestion Pipeline

High-throughput Go pipeline for data ingestion. Lockless architecture for maximum throughput.

Python + PyTorch Core

The 63-layer DAG runs in PyTorch. NumPy + SciPy for matrix exponentials. Symplectic integrators preserve phase-space volume.

About

Independent. Self-Funded.

Kairos Signal is independently built and self-funded, zero VC. The entire 63-layer ByteDAG, the ClickHouse telemetry engine, and all data pipelines are built from scratch on a self-owned infrastructure.

Every dollar of revenue goes back into the pipeline. Every signal is hashed. Every prediction is auditable. This is a machine learning research product — not an investment advisor, not a broker-dealer.

The math is real. The signals are verifiable. Every prediction is hashed into a SHA-256 chain in ClickHouse — you can audit the results yourself.

$0
VC Funding Taken
47
Assets Tracked
3,235
Counties Covered
100%
Independently Owned

Get Started

Start with the free tier. Pull your first API response. Verify the SHA-256 chain yourself. Three tiers: Free, Pro ($299/mo), and Quant ($499/mo).

Create Free Account → Download Enterprise Deck
The Math

The Math

47 peer-reviewed mathematical sources underpin the 63-layer architecture: symplectic geometry, gauge theory, and topological data analysis. Every signal is cryptographically chained via SHA-256.

LAYER 52
Doléans-Dade Exponential
Υt = exp(−∫0t βs dWs − ½∫0t ‖βs‖² ds)

The change-of-measure exponential that removes non-Markovian dependence from the BSDE. Maps the stochastic solution onto a Markovian PDE — the mathematical backbone of the stochastic adapter stack.

Reference: Doléans-Dade (1970) — implemented in PyTorch

LAYER 62
Wilson Loop — Gauge Arbitrage
W(C) = Tr(P exp(∮C Aμ dxμ)) / n

Non-trivial holonomy W(C) ≠ 1 detects topological arbitrage. The gauge connection A is derived from the log-covariance of the asset manifold — the terminal execution head of the entire DAG.

Reference: Yang & Mills (1954) — implemented via scipy.linalg.expm

LAYERS 18–27
Green's Function Convolution
(K(ℓ)v)(x) = C(ℓ) ∫D ΦL(x,y) v(y) dy

The singular kernel ΦL is hard-coded — not learned — encoding the elliptic PDE structure. Only the coefficient matrix C(ℓ) trains. This factoring yields polynomial (not exponential) parameter scaling.

Reference: Furuya & Kratsios, arXiv:2410.14788

GLOBAL INVARIANT
DAG Acyclicity — Runtime Verified
∀ v < 63, ¬CanReach v v
|V| = 63, |E| = 191

No directed cycles exist in the 63-node, 191-edge computation graph. Verified at runtime via topological sort. Every layer reachable from root.

Property: DAG acyclicity verified at runtime — no directed cycles in the 63-node graph

LAYER 57
Choquet Integral — Possibilistic Scoring
∫ f dμ = Σi (f(σi) − f(σi+1)) · μ({σ1,…,σi})

The Choquet integral scores actions under a non-additive possibility measure. Ensures the inferential model never favors actions the oracle doesn't — the mathematical foundation of the XGBoost veto gate.

Reference: Choquet (1954) — implemented in NumPy

45 LAYERS
Clarke Subdifferential ∂C
∂C f(x) = conv{lim ∇f(xk) : xk → x}

The generalized gradient at non-smooth points returns the convex hull of one-sided limits. Preserves sharp regime boundaries that smooth activations destroy — used in 45 of 63 layers.

Reference: Clarke (1990) — implemented via PyTorch autograd

§1 — Topological Data Analysis & Persistent Homology

Detects when the shape of the DePIN order book tears apart, triggering the f118 phase shift.

[1] Carlsson, G. (2009). Topology and Data. Bulletin of the AMS.
[2] Edelsbrunner, H. & Harer, J. (2010). Computational Topology. AMS.
[3] Zomorodian, A. & Carlsson, G. (2005). Computing Persistent Homology. Discrete & Comp. Geom.
[4] Bubenik, P. (2015). Statistical TDA using Persistence Landscapes. JMLR.
[5] Ghrist, R. (2008). Barcodes: The persistent topology of data. Bulletin of the AMS.
[6] Cohen-Steiner, D. et al. (2007). Stability of Persistence Diagrams. Discrete & Comp. Geom.

§2 — Symplectic Geometry & Hamiltonian Phase Space

Maps energy conservation, forcing 3072D embeddings to obey momentum and position laws.

[7] Arnold, V. I. (1989). Mathematical Methods of Classical Mechanics. Springer.
[8] Marsden, J. E. & Ratiu, T. S. (1999). Introduction to Mechanics and Symmetry. Springer.
[9] Stochastic Port-Hamiltonian Neural Networks. arXiv:2603.10078
[10] Adaptive Meta-Learning Stochastic Gradient HMC. arXiv:2604.25710
[11] Greydanus, S. et al. (2019). Hamiltonian Neural Networks. NeurIPS.
[12] Chen, T. et al. (2018). Neural SDEs. NeurIPS. arXiv:1806.07366
[13] McDuff, D. & Salamon, D. (2017). Introduction to Symplectic Topology. Oxford.
[14] Hofer, H. & Zehnder, E. (1994). Symplectic Invariants and Hamiltonian Dynamics. Birkhäuser.

§3 — Causal Inference & Exact Functional ANOVA

Vetos spurious correlation and isolates deterministic alpha inside filtering nodes.

[15] Pearl, J. (2009). Causality. 2nd Ed. Cambridge.
[16] Hooker, G. (2004). Discovering Additive Structure in Black Box Functions. KDD.
[17] Sobol, I. M. (2001). Global sensitivity indices. Math. & Comp. in Simulation.
[18] Spirtes, P. et al. (2000). Causation, Prediction, and Search. MIT Press.
[19] Peters, J. et al. (2017). Elements of Causal Inference. MIT Press.
[20] Owen, A. B. (2013). Generalized Sobol' indices. SIAM/ASA J. Uncert. Quant.

§4 — Chaos Theory & Dynamical Systems

Governs Lyapunov chaos detection and hidden market attractor reconstruction.

[21] Takens, F. (1981). Detecting strange attractors in turbulence.
[22] Eckmann, J. P. & Ruelle, D. (1985). Ergodic theory of chaos. Rev. Mod. Phys.
[23] Wolf, A. et al. (1985). Determining Lyapunov exponents. Physica D.
[24] Kantz, H. & Schreiber, T. (2004). Nonlinear Time Series Analysis. Cambridge.
[25] Packard, N. H. et al. (1980). Geometry from a Time Series. Phys. Rev. Lett.

§5 — Non-Markovian Dynamics & Rough Path Theory

Long-range memory kernels — signals survive when standard Markov random walks fail.

[26] Mandelbrot, B. B. & Van Ness, J. W. (1968). Fractional Brownian motions. SIAM Review.
[27] Lyons, T. (1998). Differential equations driven by rough signals. Rev. Mat. Iberoamericana.
[28] Friz, P. K. & Hairer, M. (2014). A Course on Rough Paths. Springer.
[29] Gatheral, A. et al. (2018). Volatility is rough. Quantitative Finance.
[30] Biagini, F. et al. (2008). Stochastic Calculus for fBM. Springer.

§6 — Gauge Theory & Non-Commutative Geometry

Loop integrals in market microstructure reveal deep structural mispricings.

[31] Yang, C. N. & Mills, R. L. (1954). Conservation of Isotopic Spin. Physical Review.
[32] Connes, A. (1994). Noncommutative Geometry. Academic Press.
[33] Baez, J. C. & Muniain, J. P. (1994). Gauge Fields, Knots and Gravity. World Scientific.
[34] Witten, E. (1989). QFT and the Jones polynomial. Commun. Math. Phys.
[35] Donaldson, S. K. (1983). Gauge theory and 4D topology. J. Diff. Geom.

§7 — Subdifferential Mechanics & Non-Smooth Optimization

Handles violent discontinuities during sudden regime fractures.

[36] Clarke, F. H. (1990). Optimization and Nonsmooth Analysis. SIAM.
[37] Rockafellar, R. T. (1970). Convex Analysis. Princeton.
[38] Moreau, J. J. (1962). Fonctions convexes duales. Comptes Rendus Acad. Sci.
[39] Shor, N. Z. (1985). Minimization Methods for Non-Differentiable Functions. Springer.
[40] Nesterov, Y. (2018). Lectures on Convex Optimization. Springer.

§8 — Stochastic Execution & Measure Theory

Probability collapse mechanisms, PSD certifications, and final scoring weights.

[41] Risken, H. (1989). The Fokker-Planck Equation. Springer.
[42] Doléans-Dade, C. (1970). Changement de variables. ZWVG.
[43] Choquet, G. (1954). Theory of Capacities. Annales de l'Institut Fourier.
[44] Carathéodory, C. (1911). Variabilitätsbereich der Fourierschen Konstanten. Math. Annalen.
[45] Øksendal, B. (2003). Stochastic Differential Equations. Springer.
[46] Karatzas, I. & Shreve, S. E. (1991). Brownian Motion and Stochastic Calculus. Springer.
[47] Ruelle, D. (1989). Statistical Mechanics: Rigorous Results. World Scientific.

47 peer-reviewed sources · 63-layer architecture · SHA-256 signal chain

Read the papers. Verify the hash chain. Full technical manual →