first_value(Customer_ID, true) OVER (
  PARTITION BY customer_name
  ORDER BY submitted ASC, event_id ASC
  ROWS BETWEEN UNBOUNDED PRECEDING AND UNBOUNDED FOLLOWING
)

• Use a timestamp submitted, not date()

• Add a stable tiebreaker (event_id, record id, etc.)

• true → ignores NULLs

• Explicit ROWS frame avoids Spark’s default RANGE behaviour with ties

  Option B : Use row_number() instead

  If you only need the “first” row deterministically:

  row_number() OVER (
    PARTITION BY customer_name
    ORDER BY submitted_ts ASC, event_id ASC
  )

  Then select or propagate the value from rn = 1.