{
  "schema": "anvil-serving/source-registry/v1",
  "observed_date": "2026-08-21",
  "sources": [
    {
      "url": "https://huggingface.co/Qwen/Qwen3.8-27B",
      "publisher": "Qwen",
      "evidence_type": "official model card and repository",
      "age_class": "current",
      "hardware_engine_relevance": "canonical base architecture, template, context, and license identity",
      "decision_impact": "anchored the community GGUF candidate to the official base model"
    },
    {
      "url": "https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/tree/main",
      "publisher": "Unsloth",
      "evidence_type": "original quant repository",
      "age_class": "current",
      "hardware_engine_relevance": "GGUF weights, MTP head, and vision projector used by llama.cpp",
      "decision_impact": "selected Q4_0 candidate artifacts"
    },
    {
      "url": "https://huggingface.co/bartowski/Qwen3.8-27B-GGUF",
      "publisher": "bartowski",
      "evidence_type": "original quant repository",
      "age_class": "current",
      "hardware_engine_relevance": "conventional GGUF Q6_K control",
      "decision_impact": "provided exact Q6_K size for the feasibility screen"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/releases",
      "publisher": "llama.cpp",
      "evidence_type": "official runtime releases",
      "age_class": "current",
      "hardware_engine_relevance": "CUDA server runtime with Blackwell support",
      "decision_impact": "selected digest-pinned b10548 runtime"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/blob/master/tools/server/README.md",
      "publisher": "llama.cpp",
      "evidence_type": "official runtime documentation",
      "age_class": "current",
      "hardware_engine_relevance": "OpenAI-compatible server, cache, metrics, slots, multimodal",
      "decision_impact": "defined launch and observability controls"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/blob/master/docs/docker.md",
      "publisher": "llama.cpp",
      "evidence_type": "official runtime documentation",
      "age_class": "current",
      "hardware_engine_relevance": "CUDA container execution",
      "decision_impact": "supported reproducible managed container recipe"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/blob/master/docs/speculative.md",
      "publisher": "llama.cpp",
      "evidence_type": "official runtime documentation",
      "age_class": "current",
      "hardware_engine_relevance": "draft-model and MTP speculative decoding",
      "decision_impact": "selected draft-MTP controls"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/issues/27296",
      "publisher": "llama.cpp community",
      "evidence_type": "current issue report",
      "age_class": "current",
      "hardware_engine_relevance": "long-context state contamination risk",
      "decision_impact": "required post-envelope functional rerun without restart"
    },
    {
      "url": "https://github.com/ggml-org/llama.cpp/issues/27282",
      "publisher": "llama.cpp community",
      "evidence_type": "current issue report",
      "age_class": "current",
      "hardware_engine_relevance": "Qwen3.8 llama.cpp compatibility tracking",
      "decision_impact": "retained runtime caveat and log review gate"
    }
  ]
}