research_task:
  description: >
    You will be given {data}, a 2D Python List[List[str|None]] structured as follows:
      - Row 0 is the header: the first column is always 'String', and the rest are language names.
      - Rows 1…n contain translation data for each key.
    YOU MUST identify the list of language columns **explicitly** from the header row for translating later.
    ONLY and must translate into languages that are actually present — do NOT assume or invent or copy (for example copy English phase to French language).

    Your responsibilities:
      1. Parse the headers to identify all **existing** language columns.
        → DO NOT assume the existence of any column unless explicitly present.
        → DO NOT create new columns (e.g., 'English').

      2. For each row:
        a. Derive the English phrase from the 'String' key:
          - Remove the prefix 'STR_'
          - Replace all underscores with spaces
          - Convert the phrase to Title Case
        b. Use this derived phrase as the translation base.
        c. IMPORTANT STEP: For each language cell:
          - If the cell is:
              • Empty  
              • Null  
              • Whitespace  
              • **OR exactly matches `english_phrase`** (**CRITICAL**: this is not a valid translation!)
              → Then translate `english_phrase` into the TARGET LANGUAGE.
              → The translation must:
                • Be natural and fluent  
                • Match Title Case  
                • Contain **no** extra punctuation, quotes, or added words  
          - Otherwise: leave the cell unchanged.
  expected_output: >
     A 2D Python `List[List[str]]` of identical shape where:
      - All originally missing translation cells are now correctly filled
      - Table structure is preserved exactly (no added columns, no reordering)
      - Existing non-empty translations remain unchanged
      - All derived English phrases are used strictly as translation bases
      - All new translations follow capitalization and output rules strictly
  agent: translator_researcher