{
  "title": "Mushroom tray and harvest-stage data for AI",
  "source_page": "https://fiund.com/industries/mushroom-flush-selection-data",
  "authorship": "Original fictional Fiund editorial exercise. No actual operating records or third-party source assets are included.",
  "example": {
    "title": "A reviewed cap does not assign the same stage to its whole tray",
    "context": "An original fictional tray sequence uses cap IDs and reviewer labels. It includes no real imagery, substrate recipe, production incident or harvesting instruction.",
    "records": [
      [
        "MU1",
        "Tray A, first view",
        "Caps C1 and C2 visible and similar in color; C3 partly occluded. Reviewer marks C1 stage S under an original practice rubric; C2 and C3 unreviewed."
      ],
      [
        "MU2",
        "Tray A, second view",
        "Fixed camera, later time. Cross-view annotation confirms C1 identity. C2 overlaps C3 and their individual boundaries are ambiguous."
      ],
      [
        "MU3",
        "Stage summary draft",
        "Draft labels every cap in Tray A as S because C1 was reviewed. Separate Tray B has no reviewed labels."
      ]
    ],
    "question": "Which stage label is supported, and what uncertainty should be retained for the other caps and evaluation tray?",
    "answer": "MU1 supports stage S for C1 only, with MU2 preserving its cross-view identity. Similar color does not review C2, and occlusion limits C3. MU2 adds a boundary ambiguity between those two caps rather than resolving it. MU3's tray-wide assignment exceeds the supplied labels. Keep object-level review and visibility explicit, and use separately reviewed Tray B as a potential holdout instead of assuming the entire tray shares one stage.",
    "failureModes": [
      "Propagating one reviewed specimen's stage to every cap in a tray.",
      "Merging overlapping specimens or assigning identities without sufficient cross-view evidence."
    ],
    "fields": [
      [
        "Tray, cap and view timestamp",
        "Preserves individual identity."
      ],
      [
        "Visible boundary and occlusion",
        "Defines what can be inspected."
      ],
      [
        "Rubric and reviewer-label scope",
        "Limits the supported stage assignment."
      ]
    ],
    "holdout": "Use an original second-tray sequence with independently reviewed caps and changed spacing. Evaluate transfer without adjacent-frame overlap with development data.",
    "boundary": "Use owned demo trays or synthetic annotations. Clear hygienic-area access separately and exclude formulations, plans and employee details."
  }
}