{
  "timestamp_utc": "2026-10-15T05:39:25.526544+00:00",
  "completed_before_resume": 313,
  "fixes": [
    "Summary handles ChatClient tool_calls lists as well as agent integer counts.",
    "At the six-tool budget, finalization omits tool definitions and explicitly requests an evidence-only final answer.",
    "Summary and HTML costs include every billed audit response, including failed finalization attempts."
  ],
  "affected": [
    {
      "question_id": "Q006",
      "model": "google/gemini-3.8-flash",
      "strategy": "lancedb/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q055",
      "model": "google/gemini-3.8-flash",
      "strategy": "qdrant/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q068",
      "model": "google/gemini-3.8-flash",
      "strategy": "lancedb/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q057",
      "model": "google/gemini-3.8-flash",
      "strategy": "qdrant/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q088",
      "model": "google/gemini-3.8-flash",
      "strategy": "qdrant/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q057",
      "model": "google/gemini-3.8-flash",
      "strategy": "lancedb/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    },
    {
      "question_id": "Q068",
      "model": "google/gemini-3.8-flash",
      "strategy": "qdrant/agentic",
      "error": "Model ignored final-answer instruction after tool budget"
    }
  ],
  "unchanged": "Completed results are retained; no extra tool executions are permitted during recovery."
}