{
  "title": "Driving school simulator feedback data for AI",
  "source_page": "https://fiund.com/industries/driving-school-simulator-feedback-data",
  "authorship": "Original fictional Fiund editorial exercise. No actual operating records or third-party source assets are included.",
  "example": {
    "title": "A debrief cue belongs to a run, not just a matching timestamp",
    "context": "Two fabricated simulator runs use abstract event labels. This is a detached attribution exercise, with no driving procedure, licensing judgment or real learner information.",
    "records": [
      [
        "DS1",
        "Practice run R1 timeline",
        "Scenario Blue; cue C appears at 00:20."
      ],
      [
        "DS2",
        "Instructor note",
        "Explicitly names R1 / Blue / 00:20 and records exercise observation X."
      ],
      [
        "DS3",
        "Practice run R2 timeline",
        "Scenario Amber; cue C appears at 00:20 too. An unlabeled note says X at 00:20 without run or scenario."
      ]
    ],
    "question": "Which run has an attributable instructor observation, and can the unlabeled note establish the same result for R2?",
    "answer": "DS2 explicitly joins the observation to R1, scenario Blue and cue time 00:20 in DS1. R2 shares a cue label and time but has a different run and scenario. The unlabeled note in DS3 lacks those identifiers and cannot establish an R2 observation. Flag it for instructor review rather than copying the R1 finding across runs or using it to judge readiness.",
    "failureModes": [
      "Matching instructor notes using elapsed time alone.",
      "Treating recurring cue C as a unique event across all scenarios."
    ],
    "fields": [
      [
        "Run and scenario ID",
        "Defines the source session for feedback."
      ],
      [
        "Event time and cue",
        "Locates the observation within the identified run."
      ],
      [
        "Attribution confidence and reviewer",
        "Makes an ambiguous note a review item rather than a result."
      ]
    ],
    "holdout": "Use R2's cue C at 00:25 and attach a verified note referencing that later time. Test whether the model aligns it to R2 without importing the earlier R1 timestamp.",
    "boundary": "Use synthetic timelines only. Exclude real routes, students, vehicle IDs, road video, device data and safety incidents."
  }
}