Five model artifacts in the official Ollama library solve zero of 164 tasks.

Executing 327 quantized code-capable artifacts (305 from the official Ollama library across 15 model lines, 22 from top HuggingFace community repos) through a calibrated 15-task smoke suite found five silently defective: four Qwen2.5-Coder-3B conversions and one phi3.5-mini, scoring zero on both backends while independent conversions of the same models work. That's 1.6% of official artifacts. Two produce output whose surface statistics sit inside the healthy range, so nothing short of execution catches them. arXiv 2609.05881 The released quantcheck tool is the acceptance gate model registries currently lack.