You will receive an initial excerpt from a document (concatenated text from the first ~200 dataframe indices, already cleaned).

Goal: return **only** a short title for the document.

Priority (in this order):
1) If the excerpt already contains a clear cover title or a repeated header (e.g., same header across the first pages), use that exact title.
2) If there is a common header across the initial pages that looks like the document title, use it.
3) If no explicit title is found, generate a concise, descriptive title based on the excerpt.

Rules:
- Respond with the title only, no quotes and no extra text.
- If multiple variants exist, pick the most representative and clean one (strip page numbers/repetitive boilerplate).
- Prefer to keep the input language.

---------------- Excerpt ----------------
{{ document_excerpt }}
----------------------------------------
