Show the research flow. Protect the implementation.
This page explains the major scientific modules and evidence gates at a conceptual level. It intentionally does not disclose proprietary feature construction, policy internals, execution calibration, private source layout, model artifacts, credentials or operational infrastructure.
Five layers, one controlled evidence path.
Arrows show scientific dependency, not a production network diagram.
Historical market data
Time-ordered source data enters a causal research pipeline under strict chronology and provenance controls.
CAUSAL INPUTCausal observations
Market evidence is transformed into model-ready observations while preserving information availability and frozen schema rules.
PRIVATE RECIPEFrozen candidate generation
Preregistered experiments train candidate policies under fixed budgets and fixed treatment definitions.
NO TRAIN SELECTIONExecution + risk environment
Actions are evaluated inside a futures-aware accounting and risk environment with causal execution semantics.
PRIVATE CALIBRATIONFresh → Selection → TEST
Performance evidence is revealed only after complete preregistered evaluation and separate authorization gates.
FAIL-CLOSEDWhat we disclose — and what we do not.
The goal is scientific transparency without publishing the recipe for reproducing or attacking the system.
Safe to disclose
- Research stages, experiment IDs and high-level study purpose.
- Evidence boundaries: TRAIN, Fresh Validation, Historical TEST and deployment.
- Frozen status, aggregate structural counts and CI pass/fail state.
- Whether selection, TEST access or live trading is authorized.
- High-level causality, preregistration and fail-closed methodology.
Kept private
- Exact feature-generation recipes, transformations and proprietary signals.
- Internal model weights, checkpoints, optimizer state and training artifacts.
- Detailed execution calibration, private risk tuning and operational parameters.
- Private repository structure beyond what is already intentionally published.
- Credentials, infrastructure topology, deployment endpoints and exchange access.
The public map emphasizes gates, not alpha.
Causality gate
Information enters the model only when economically available.
Freeze gate
Treatments, candidates and decision rules are locked before Fresh reveal.
Reveal gate
Partial Fresh performance has no selection authority.
Deployment gate
Research success does not automatically authorize live capital.