Task: From DOCUMENT TEXT below (plain-text legal document extract), extract a row-by-row consistency dictionary for {source_lang_name} → {target_lang} ({prompt_target_phrase}).

Each content segment is translated in isolation without surrounding context. Output ONLY source→target mappings for terms where an isolated pass would likely (1) leave source-language text untranslated, (2) render the same phrase differently across segments, or (3) choose a plausible but wrong target-language equivalent.

Do NOT extract writing-style rules. Do NOT reproduce definitional sections or term lists verbatim. Do NOT build a document encyclopedia.

OUTPUT FORMAT — begin your reply with this exact line:
{style_full_marker}

Then two blocks of newline-delimited dictionary lines only (no writing style):

{style_full_marker}
<source> → <target>
(15–35 lines; fewer high-signal entries is better)

{style_compact_marker}
<source> → <target>
(8–15 lines; include every quoted technical term)

{dictionary_format_rules}

WHAT TO INCLUDE (only if confusing in row-by-row translation):
- Quoted technical or device terms — bare term as source, one locked target (highest priority)
- Legal or regulatory structural labels with one fixed target-language form
- Administrative or procedural phrases with non-obvious or unstable target equivalents
- Defined terms whose target rendering varies segment to segment
- Procedural codes, suffixes, or identifiers that need a short gloss
- Document-specific acronyms only when expansion is non-obvious and reused

WHAT TO EXCLUDE:
- Long definitional phrases or wrapper text around a quoted term (extract the term itself when possible; row matching is flexible but bare terms work best)
- Structural citation labels already governed by corporate/language rules (e.g. art., Art., parágrafo único) unless the document uses a non-standard form

{terminology_rules}

LIMITS (excess output is discarded automatically):
- FULL block: ≤35 lines and ≤1,800 characters
- COMPACT block: ≤15 lines and ≤700 characters
{chunk_note}DOCUMENT TEXT:
{text_content}
