Explain a route | LLMRead the routing library's explanation of a catalog, and see why a candidate was refused.LLMhow-tollmhow-todeveloperoperatorbuild-agent-systemsdeploy-operate-productsreference

Explain a route

cargo run --locked -p b10x-llm-routing --example explain

The example reads examples/catalog.toml, which declares a self-hosted provider with one anonymous local endpoint and one authenticated remote endpoint, and a coding route whose two targets are ordered.

The output

{
  "route_id": "coding",
  "alias": "code",
  "config_digest": "1365b05fe63542f8211e21a9a2566c71f2a6d44ec9bdc564ba46c592a861825f",
  "input_tokens": 512,
  "selected_target_id": "coding-primary",
  "targets": [
    {
      "target_id": "coding-primary",
      "serving_model_id": "local-small",
      "position": 0,
      "provenance": {
        "protocol": "chat-completions",
        "provider": "my-lab",
        "account": "local",
        "endpoint": "local-vllm",
        "model": "small",
        "binding_revision": "1a6ea40f98c8eef26ff66e7c753dda816f45a1aa3488bb31f62ef25a89c328a1"
      },
      "auth_kind": "anonymous",
      "billing_kind": "self-hosted",
      "capabilities": {
        "tools": false,
        "tool_choice": false,
        "temperature": false,
        "top_p": false,
        "reasoning_efforts": [],
        "context_window": 4096,
        "max_output_tokens": 1024
      },
      "rejections": []
    },
    {
      "target_id": "coding-secondary",
      "serving_model_id": "remote-large",
      "position": 1,
      "provenance": {
        "protocol": "responses",
        "provider": "my-lab",
        "account": "remote",
        "endpoint": "remote-models",
        "model": "large",
        "binding_revision": "5dd662a7d1b9e5280a10e6b691ff04c79326d1562d18d3206e9d1bd141ed9a13"
      },
      "auth_kind": "bearer",
      "billing_kind": "metered",
      "capabilities": {
        "tools": true,
        "tool_choice": true,
        "temperature": true,
        "top_p": true,
        "reasoning_efforts": ["medium", "high"],
        "context_window": 32768,
        "max_output_tokens": 8192
      },
      "rejections": ["fallback-disabled"]
    }
  ]
}

How to read it

config_digest identifies the validated catalog. Two equivalent files that differ only in row order or decimal spelling share a digest; changing a declared fact changes it.
binding_revision hashes the complete validated binding, including the endpoint URL, upstream model, auth reference and capabilities. It is the sixth coordinate that opaque continuation state is checked against, so repointing local-vllm at another URL invalidates prior opaque state — deliberately.
rejections: ["fallback-disabled"] is the second target's only reason for being ineligible. The route sets fallback_enabled = false, and omission is false. Turn it on and the same target becomes a named alternative.
auth_kind and billing_kind are independent. anonymous + self-hosted and bearer + metered are both ordinary bindings.

What the explanation is not

It resolved no secret — lab-llm-token appears in the catalog as a reference and nowhere in the output. It sent no request, provisioned nothing and contains no prompt or opaque payload. Listing or explaining routes must never do any of those things.

The output​

How to read it​

What the explanation is not​

The output

How to read it

What the explanation is not