{
  "title": "Compare effort before a higher-effort retry",
  "caption": "Offline evaluations use the same acceptance rule. Online verification may cause a supported retry at higher effort if the remaining budget is sufficient. This does not guarantee success. Failed attempts add cost and latency. Success and no-budget cases are constructed examples. The author made this explanatory diagram with example playback, and animation time is not measured runtime.",
  "alt": "Evaluation flow: compare supported efforts offline. Online verification can cause a retry at higher effort if the task budget is sufficient.",
  "sources": [
    {
      "path": "content/blog/2026-07-22_reasoning-effort-inference-systems/post.md",
      "description": "Article explanation and retained source citations"
    }
  ],
  "limitations": [
    "The execution is an example. Playback duration is not measured runtime.",
    "Constructed acceptance outcomes are not benchmark measurements. A supported higher mode and remaining budget are necessary for a retry. A higher mode can fail the rule."
  ]
}