You write one realistic input item for a decision application.

Application: {{APPLICATION}}
Item: {{ITEM_KIND}}, about {{LENGTH}}.
{{ENTITIES}}

The application will ask these questions about the item. Write the item so that a careful,
fair reader would reach the intended reading where one is given:
{{TARGETS}}

How to realize the intended readings:
- "clear": the item contains evidence that settles the question for a careful reader, though it
  may take reading several parts of the item or combining cues; never state the answer in the
  question's own words.
- "borderline, mostly X": X is the best reading, but the evidence also genuinely supports one other
  option that a careful reader could confuse with X (you choose it); the remaining options are ruled out.
- a probability for yes/no: the evidence should make that probability reasonable. 50% means the
  relevant evidence is truly balanced or missing, not that the item ignores the question.
- {{DIFFICULTY}}

Let the evidence carry each reading: never write the conclusion itself (for instance that something
is urgent, approved, a violation, or that someone is or is not in some category).

The item must read like a real artifact of its kind: concrete details, natural voice, realistic
noise and irrelevant details, no headings that mirror the questions, no mention of questions,
answers, labels, probabilities, or of being generated. Facts must stay consistent across the item.

Output strict JSON only: {"state": "the item text"}
