{
  "title": "Before you use feedback from a run, examine the run.",
  "caption": "Traces and evaluation connect data, training, runtime and product behavior. Validated feedback can supply input for curation and a new run with explicit training, with source history kept. This is not automatic online learning. The author made this explanatory diagram. The playback is an example, and animation time is not measured runtime.",
  "alt": "Feedback flow: training uses versioned data. Trained weights are installed in a runtime. Traces and evaluation send validated results to curation and a new training run.",
  "sources": [
    {
      "path": "content/blog/2026-07-13_ai-infrastructure-is-not-plumbing/post.md",
      "description": "Article explanation and retained source citations"
    }
  ],
  "limitations": [
    "The playback is an example. Playback time is not measured runtime.",
    "This loop must use explicit training. Traces and product failures do not automatically update weights."
  ]
}