A Verification Ladder for Low-Cost AI Coding Models

I used to ask a simple question when routing coding tasks across models: Which model is strong...