ATRX v4.0 is organised as seven architectural pillars, executed by a five agent decision pipeline, linked to subscriber terminals through a schema signed transport layer, and refined continuously by a post trade learning loop. The summary describes the shape of the system at a level sufficient to evaluate the work. Proprietary edge lives in the configuration of the components.
Seven pillars. Five agents. One veto.
THE SEVEN PILLARS
01 · Foundation
Data integrity
Temporally aligned, source signed, corrected tick and macro series. Nothing enters without provenance and timestamp discipline.
02 · Hypothesis
Causal thesis layer
Every admitted strategy carries a stated transmission mechanism rooted in flow, macro, carry, or microstructure. Correlation alone does not qualify.
03 · Inference
DNA alpha factory
A structural multi factor research engine. Strategy candidates are composed from building blocks, mutated under constraints, and admitted only by walk forward evidence.
04 · Arbitration
Multi tier intelligence
A layered model of language and reasoning systems validates each candidate under directional and structural constraints. The arbiter never overrides the quantitative thesis.
05 · Context
Structural regime engine
State estimation, density modelling, and a meta classifier identify the prevailing regime and gate which strategies are permitted to act.
06 · Constraint
Risk gatekeeper
Anti martingale framework with drawdown tiered position sizing, correlation tracking, circuit breakers, and absolute veto authority.
07 · Execution
Cross platform connectivity
Bridges and connections to multiple trading systems and broker venues, with broker agnostic order management and a signed trade ledger.
DECISION PIPELINE
Each decision flows through research, model ensemble, arbiter, risk, and execution in series. The risk gatekeeper receives a mandatory handoff. Its feedback updates both the immediate decision and the regression set used for retraining.
TRANSPORT LAYER
Subscribers install a local agent that speaks a schema signed protocol to the ATRX intelligence cloud. Instructions are signed, schema validated, and replay protected. No credentials leave the subscriber's environment.
POST TRADE LEARNING
Closed trades feed a signed ledger that backs a walk forward regression set. Adapter updates are proposed against that set and promoted only after risk gatekeeper review. Retrains are never silent. Data drift is monitored and flagged.
Engineering Footprint
The platform is implemented end to end in production grade Python across a multi thousand line research engine, a high throughput runtime, durable state, compiled inference, asynchronous transport, and a TypeScript operations dashboard. Bridges and connections to multiple trading systems are maintained for broker agnostic deployment.
In Development
The ATRX Agent extends autonomous execution into autonomous research, proposing candidates that pass through the same validation gates as human authored work. An in house AI system is being engineered to bring the full intelligence stack under the firm's own roof.