Σ
Adaptive Alpha
Long- and short-horizon signal systems
Rolling models read recent regimes instead of treating market structure as static. Multiple mechanisms, one rigorous evidence language.
AI-Native Market Systems / Hong Kong / US
Markets areliving systems.
We build adaptive intelligence for them — grounded in mathematics, with long- and short-horizon signals, execution-aware learning, and autonomous research systems governed by machine-speed risk controls.
Through the window of differential equations, one glimpses the light of the real world.
Nominal
Nominal
Nominal
Signal intelligence
Adaptive, long- and short-horizon, multi-mechanism
Execution science
Market-aware from replay to live
Machine governance
Risk veto remains absolute
Evidence memory
Every decision leaves a trace
Neo Lab / Research Index
Multiple intelligences.One disciplined system.
Our edge is not a single model. It is a research machine that discovers, falsifies, executes, and remembers — across strategies designed to fail at different times.
↝
Execution Science
Research that survives contact
Signals are studied against the prices, frictions, timing, and failure modes they will face in the market — not in an idealized vacuum.
N
Portfolio Intelligence
Independent strategy constellations
More than 100 independent sleeves, assembled for distinct mechanisms and distinct failure times — the opposite of one monolithic bet.
ψ
Neural Market Models
Adaptive representations of market state
Dynamic and recurrent models study how useful state can persist, update, and reset across changing regimes without treating market structure as stationary.
π
Adaptive Execution
Machine-learned execution policies
A broader execution intelligence program spanning supervised models, online adaptation, and reinforcement learning for placement, timing, and capital efficiency.
⌁
Market Microstructure
US–HK and co-location intelligence
Cross-session market memory, venue-aware architecture, and latency research for a future where geography becomes another model input.
Signals learn.Risk does not deform.
A fiber bundle with a learned connection. The base manifold is the market. The fibers are feature space. The connection — how signals are transported — is learned. One section of the bundle is never allowed to deform: the veto.
Regimes deform the manifold. The invariants survive.
Evidence beforeoutcomes.
Every decision leaves a cryptographic trace before the market answers. Research may explore. Execution accepts only what survives verification.
RESEARCH MEMORY
Decisions precede outcomes
Chronology is part of truth. Evidence is sealed before consequence can rewrite it.
Trust is continuously earned
Every boundary re-establishes provenance. Uncertain evidence has no authority.
The lab remembers
Success, failure, and regimes become durable context without rewriting history.
Memory is infrastructure.
Forward ↔ Inverse
The outcome is visible.The mechanism is not.
AI learning and quantitative inference both work backward from observed effects. AI learns a model from data; quant research infers candidate market mechanisms from price, volume, order flow, and context. Without constraints, the answer is rarely unique or stable.
y = F(θ, x₀) + ε
Several mechanisms can leave the same footprint.
F
Forward problem
Given the process, compute the result
Choose a mechanism, initial state, and evolution rule; then simulate the observations it would produce.
F⁻¹
Inverse problem
Given the result, infer the process
Start with the recorded tape and recover a candidate latent state, operator, or mechanism that could explain it.
λ
Regularization
Discipline makes inference usable
Point-in-time data, simplicity, trading costs, hostile regimes, and forward evidence constrain the many explanations that fit history.
PDE/ODE, state-space models, stochastic processes, and neural networks can define candidate forward operators. None proves that the market obeys one equation; each remains a hypothesis under test.
The systemwatches itself.
State, decisions, orders, and risk move through one observable surface. The machine is never allowed to become a black box.
dnalyaw-lab :: synthetic-system-surface
OBSERVING· CROSS-SESSION
Signal Coherence+0.684σcross-session · normalized state
State Evolution
OBSERVEADAPTREMEMBER
Active ContextsMULTIindependent research loops
Evidence StateATTESTEDprovenance attached
Latent Liquidity Field · Venue Composite
state divergence: bounded
Decision Traces
TIMESYMSIDEQTYPRICESTATUSID
T+00:07VECTOR-AALIGN0.820.74FILLED#A1
T+00:09VECTOR-BOFFSET0.460.68FILLED#B7
T+00:11VECTOR-CALIGN0.610.71PARTIAL#C3
T+00:14VECTOR-DOFFSET0.290.65WORKING#D2
T+00:16VECTOR-EALIGN0.530.77WORKING#E9
MemoryDURABLE
SurfaceMULTI-VENUE
StateOBSERVABLE
Measureeverything.
Execution is half the alpha. Every fill becomes evidence: arrival, path, friction, impact, and the tails that averages try to hide.
Latency Histogram
Slippage Distribution
Evidence Coverage
Decision to fillTraceable
Tail behaviorObserved
Venue contextAttached
Replay stateReproducible
Markets cross borders.So does intelligence.
A cross-session research fabric spanning the United States and Hong Kong — learning from one market while the other sleeps.
NEW YORK / SIGNAL ORIGIN
US market structure, execution research, and long- and short-horizon intelligence.
HONG KONG / SESSION MEMORY
Asian session context, cross-market hypotheses, and a laboratory where market behaviour is structured differently.
ONE RESEARCH SURFACE
Different venues and clocks, normalized into evidence the research system can compare.
Selected Writing
The bridge is mathematical.Each field keeps its own proof.
Selected work showing what connects mathematics, neural networks, and quantitative research — and where each domain keeps its own burden of proof.
Mathematical foundationA Question Chern Asked, Answered by Two Worlds Numerical manifolds and neural networks arrive at the same answer, thirty-seven years apart.Neural-network foundationThe Four Realms of Neural Networks From PDE solvers to manifold geometry and measurement — a framework for thinking about learned representations.Quantitative researchThe Geometry of Alpha Why the durable quant moat is a factory that keeps producing and validating signals, not a single strategy recipe.Research systemsDnalyaw, Engineered from Scratch Polyglot Rust/Go/Python architecture, hundreds of features, and disciplined risk — how vertical integration of research and execution becomes the real edge.