Five foundation articles for evidence-backed AI, MQL5, and open-source work.
A practical workflow that starts from an entry point, gathers bounded context, checks blast radius, and separates static evidence from compiler and runtime truth.
AI can read many lines of code and still lose its bearings. A code graph connects symbols, calls, and event handlers to explicit evidence.
Understand the local MCP boundary, register mql5-codegraph-mcp per client, select a separate trusted project root, and maintain honest RAM-only snapshots.
A Windows guide from Python 3.11 and the public wheel to a trusted MQL5 repository, a first JSON snapshot, and honest result checks.
Keep AI-assisted creative speed while adding specs, small changes, impact review, tests, and a journal so the project remains understandable months later.