llm-generated-complex-code-contains-systematic-logic-bugs
Explicit syntax review of a 480-line LLM-generated strategy file found 4 critical logic bugs that would cause silent failures in production: unbounded position sizing on low-ATR symbols, a backward-moving trailing stop, entries in the last trading minute triggering immediate EOD close, and invalid alert syntax specific to strategy mode. These are not caught by a compiler — only by deliberate human-equivalent logic review. Any code generation task >300 lines should include a dedicated logic-review pass before delivery.