Skip to content

architecture.md

Latest commit

 

History

History
103 lines (79 loc) · 3.13 KB

architecture.md

File metadata and controls

103 lines (79 loc) · 3.13 KB

Architecture Overview

This file describes the architecture that is currently implemented in code, and also calls out the planned architecture that still exists only as design intent.

Current runtime architecture

flowchart LR
    UI["React + Vite frontend"] --> API["FastAPI backend"]
    UI --> WS["WebSocket run stream"]
    API --> DATA["Dataset service + DataManager"]
    API --> PY["PythonStrategyEngine"]
    API --> CMP["ComparisonEngine"]
    API --> CHAT["ChatService + OllamaClient"]
    API --> BRIDGE["BridgeService"]
    API --> PERM["PermissionManager"]
    DATA --> STORE["JSON + CSV storage under data/"]
    BRIDGE --> STORE
    PERM --> STORE
    API --> STORE
Loading

Implemented architecture

  • Frontend:
    • React + TypeScript + Vite
    • React Router for /imports, /workspace, /runs, and /settings
    • lightweight-charts for candle and overlay series rendering
  • Backend:
    • FastAPI REST endpoints
    • WebSocket stream for run progress updates
    • Local Python execution through PythonStrategyEngine
    • Manual Pine bridge artifact ingestion through BridgeService
    • Ollama integration through ChatService and OllamaClient
  • Active storage:
    • dataset CSV files
    • JSON indices and artifacts under data/cache and data/artifacts

Planned architecture that is not active yet

  • SQLite for metadata and approvals
  • DuckDB and Parquet for candle persistence and analytical artifacts
  • Automated TradingView bridge worker
  • Market-data provider ingestion for Polygon and future adapters
  • Dedicated background worker processes outside the main API process

Execution modes

local_compare

  • Uses a saved dataset from Excel or CSV
  • Runs the Python strategy locally
  • Uses an optional manual Pine bridge artifact for Pine truth
  • Best for deterministic replay and indicator debugging

pine_bridge

  • Uses a saved dataset plus an attached bridge artifact
  • Treats the uploaded Pine artifact as the source of truth for Pine values
  • Best for exact parity checks when the user exports Pine results externally

Current "live" behavior

The current live run mode is not a network-backed market feed. It replays a saved dataset incrementally:

  • warmup bars are loaded first
  • the remaining bars are appended one bar per second
  • Python outputs and comparison results are recalculated at each step
  • a WebSocket emits progress snapshots for the selected run

Approval model

  • Read access can be granted and stored in permission history
  • Write access remains explicit and short-lived
  • Chat requests with apply_fix are blocked unless Python write permission is active
  • The current frontend exposes approval toggles and status, but does not yet apply returned patches into files

Comparison contract

All engines are normalized into the same payload shape:

  • candle points
  • indicator series
  • trade events
  • mismatch details
  • warnings
  • run metadata

Comparison happens after alignment by:

  • symbol
  • timeframe
  • timezone
  • timestamp or bar index
  • tolerance

Related docs