Update to Power BI Desktop

{
  "name": "powerbi-forecast-trust-market",
  "description": "Use when designing or evaluating a Power BI forecast trust market, human override tracking, sales and finance confidence scoring, forecast accountability, and self-learning model-vs-human accuracy loops.",
  "included_files": [
    {
      "relative_path": "agents/openai.yaml",
      "size_in_bytes": 144
    }
  ],
  "skill_md_contents": "---\r\nname: powerbi-forecast-trust-market\r\ndescription: Use when designing or evaluating a Power BI forecast trust market, human override tracking, sales and finance confidence scoring, forecast accountability, and self-learning model-vs-human accuracy loops.\r\n---\r\n\r\n# Power BI Forecast Trust Market\r\n\r\nUse this skill when the user wants the forecast to learn from human judgment, overrides, and organizational accuracy over time.\r\n\r\n## Concept\r\n\r\nEvery forecast can receive confidence signals from AI, Sales, Finance, Supply Chain, and Management. Later actuals determine which source was calibrated and which source was biased.\r\n\r\n## Workflow\r\n\r\n1. For each forecast version, capture:\r\n   - AI forecast\r\n   - roll forecast\r\n   - budget\r\n   - human override\r\n   - confidence percentage\r\n   - reason code\r\n   - owner or role\r\n   - timestamp\r\n2. After actuals arrive, calculate:\r\n   - absolute error\r\n   - bias\r\n   - directional accuracy\r\n   - calibration error\r\n   - trust score by role, segment, customer, product, and horizon\r\n3. Feed the trust score back into the next forecast cycle:\r\n   - high-trust human input can challenge model output\r\n   - low-trust input remains visible but gets lower weight\r\n   - biased sources trigger review rather than automatic blending\r\n\r\n## Suggested tables\r\n\r\n- `AI_Forecast_Version`\r\n- `AI_Forecast_Override`\r\n- `AI_Forecast_Actuals`\r\n- `AI_Forecast_TrustScore`\r\n\r\n## Required metrics\r\n\r\n- WAPE\r\n- bias\r\n- hit rate within tolerance\r\n- override lift versus AI baseline\r\n- override lift versus roll forecast\r\n- confidence calibration\r\n\r\n## Guardrails\r\n\r\n- Do not turn trust scores into personal blame. Frame them as calibration by source and context.\r\n- Preserve the original AI forecast and human override separately.\r\n- Require reason codes for overrides that materially change the forecast.\r\n"
}