Concept
Learning Systems
Feedback infrastructure that transforms every outcome into improved future decisions.
Research Area
Intelligence
Status
Active Research
Version
0.2
Last Updated
July 2026
Related Discipline
Machine Learning, Cybernetics
Implementation
Orqis
Learning Systems close the loop between execution and future decision-making. Every outcome — whether successful or unsuccessful — generates information that can improve future exploration, qualification, and allocation. The goal is not just to learn what happened, but to systematically improve the process that led to the decision.
Precise Definition
Learning systems are infrastructure that transforms execution outcomes into structured institutional knowledge, injecting cycle-over-cycle improvements into generation, qualification, and allocation processes.
Key Properties
- Outcome attribution — learning begins with decomposing outcomes into parameter-level contributions, not aggregate pass/fail
- Feedback injection — learning artifacts are consumed by upstream systems (generation prompts, qualification thresholds, allocation weights) in structured, auditable ways
- Cycle-over-cycle improvement — each generation-qualification-execution cycle produces learning that measurably improves the next cycle
- Regime awareness — learning is conditioned on environmental context, preventing the system from overfitting to patterns that only hold in specific regimes
- Decay and refresh — older learning signals are weighted less as market structure evolves, with periodic retraining to prevent stale institutional knowledge
Why It Matters
Without learning systems, decision infrastructure is static — it can qualify decisions but cannot improve its own qualification over time. Learning systems are what transform decision infrastructure from a filter into a compounding advantage.
Boundary
What this concept is not
Learning systems are not machine learning models. ML models may be components within a learning system, but the learning system itself is infrastructure that compounds institutional knowledge across the full decision lifecycle.
Examples
Learning summary aggregation: weekly batch that synthesizes backtest patterns, paper trading outcomes, exit attribution, and fee-bleeder rates into structured guidance injected into AI generation prompts
Fee-bleeder detection: the 6h evaluation batch identifies strategies with average returns below transaction costs, feeding this signal back into generation filtering to suppress known unprofitable family-timeframe combinations
Indicator ranking from outcomes: the M4 indicator ranker uses Wilson-score Bayesian smoothing on closed position data to rank indicators by regime-conditioned effectiveness, guiding future strategy generation
Conceptual Neighbors
Requires
- Outcome Attribution
- Evidence Systems
Enables
- Generation improvement
- Qualification refinement
Produces
- Institutional knowledge
Opposes
- Static decision systems
Current Research Questions
- ?How should learning systems weight recent outcomes vs. historical patterns?
- ?Can meta-learning improve the learning rate of qualification systems?
- ?What is the minimum outcome volume needed for reliable learning signal?
- ?How do learning systems avoid overfitting to regime-specific patterns?
Related Frameworks
Related Research
Related Papers
This concept is actively implemented in Orqis, Warren Labs' first decision infrastructure platform.