Historical Data Engine
Acquires and normalizes OANDA historical candles, metadata, sessions, data quality signals, and replay support.
Output: Historical datasets.Architecture Reset
Axiom Intelligence Architecture
Axiom's primary purpose is to become an evidence-driven trading intelligence platform. It discovers, validates, remembers, and improves trading knowledge using historical and live market data.
Current Sprint 4 boundary: Learning Engine, Decision Engine, Intelligence Assistant, Evidence Explorer, and Trade Review vertical slice. Do not add trading signals, autonomous execution, strategy optimization, or unapproved custom algorithms.
Canonical Data Flow
Every Module Fits One Pipeline
Historical OANDA Data
Research Engine
Backtesting Engine
Learning Database
Recommendation Engine
OANDA Practice
Journal
Continuous Learning
Component Diagram
Responsibilities and Outputs
Research Engine
Generates hypotheses worth testing. It asks what recurring market behaviors, predictive features, or relationships deserve validation.
Output: Research hypotheses.Backtesting Engine
Validates hypotheses against historical OANDA data, measures outcomes, calculates statistical metrics, and eliminates weak hypotheses.
Output: Evidence, not recommendations.Learning Database
Stores tested hypotheses, performance history, statistical confidence, market conditions, timeframe and regime performance, confidence history, validation history, and retirement history.
Output: Permanent trading knowledge.Recommendation Engine
Uses Learning Database evidence, current context, similar historical situations, market regime, and risk profile to generate traceable recommendations.
Output: Evidence-based recommendations.Trade Execution
Consumes approved recommendations through trade tickets, OANDA Practice execution, position monitoring, and journal updates.
Output: Practice trade state and journal entries.Continuous Learning
Learns from live trades, historical testing, replay, user overrides, and outcomes.
Output: Updated evidence, confidence, and recommendation quality.Intelligence-First Roadmap
New Implementation Order
Architecture Principles
How Intelligence Work Is Governed
Intelligence Before Interface
Knowledge generation takes priority over UI.
Evidence Before Recommendation
Recommendations must be backed by measurable evidence.
Learn Once, Use Forever
Knowledge generated through research and backtesting becomes permanent platform knowledge.
Explainability
Every recommendation must explain why, based on what evidence, confidence, and historical support.
Continuous Improvement
Every live trade should improve the platform.
Sprint Recommendation
Sprint 4 Intelligence Foundation
Sprint 4 should define how evidence becomes permanent learning and how trade review feeds decision quality. It should not recommend or execute trades.
Recommended Deliverables
- Historical Data Engine boundary
- Research hypothesis model and service contract
- Backtesting Engine architecture proposal and test contract
- Learning Database schema proposal
- Evidence record model
Explicit Non-Goals
- No recommendation generation
- No trade execution
- No Advisor or Mission Control behavior
- No new autonomous logic
- No database migration in this reset task