Invite de synthèse par chaîne de densité

[ { "Missing_Entities": "French translation; VERBATIM placeholders; brand/model names; param flags; code/JSON keys; quoted in-image text; structure 1:1; no em-dashes; preserve diacritics; output only translation", "Denser_Summary": "This article discusses the task of translating content into French while preserving specific elements. It requires keeping VERBATIM placeholders like [X]/{x}, brand/model names, param flags such as --ar and 85mm, code/JSON keys, and quoted in-image text. The structure must be maintained 1:1, using ' - ' instead of em-dashes, and preserving diacritics. The output should only contain the translation without any preamble or fences." }, { "Missing_Entities": "generate concise summaries; entity-dense; 5 iterations; step 1 identify missing entities; step 2 write denser summary; identical length; guidelines for first summary; verbose language; fillers; ~80 words", "Denser_Summary": "The task involves translating French while preserving VERBATIM placeholders [X]/{x}, brand/model names, param flags (--ar,85mm), code/JSON keys, and quoted in-image text, maintaining 1:1 structure, using ' - ' not em-dashes, preserving diacritics, and outputting only translation. Separately, it requires generating increasingly concise, entity-dense summaries over 5 iterations: step 1 identifies 1-3 missing informative entities (delimited by ';') from the article, step 2 writes a new denser summary of identical length covering all previous and new entities. The first summary should be ~80 words, verbose, and non-specific, using fillers like 'this article discusses'." }, { "Missing_Entities": "missing entity criteria; relevant; specific (≤5 words); novel; faithful; anywhere; make every word count; rewrite for flow; make space with fusion/compression; remove uninformative phrases; self-contained; never drop entities; add fewer if needed", "Denser_Summary": "The French translation task preserves VERBATIM placeholders [X]/{x}, brand/model names, param flags (--ar,85mm), code/JSON keys, quoted in-image text, 1:1 structure, ' - ' not em-dashes, diacritics, output only translation. The summary generation process over 5 iterations: step 1 identifies 1-3 missing entities (relevant, ≤5 words, novel, faithful, anywhere; delimited by ';'), step 2 writes a denser summary of identical length. Guidelines: first summary ~80 words, verbose, non-specific; rewrite for flow; make space via fusion/compression; remove uninformative phrases like 'the article discusses'; summaries self-contained; never drop entities; add fewer if space limited." }, { "Missing_Entities": "JSON output; list of 5 dictionaries; keys 'Missing_Entities' and 'Denser_Summary'; answer in JSON; no other format", "Denser_Summary": "The French translation task preserves VERBATIM placeholders [X]/{x}, brand/model names, param flags (--ar,85mm), code/JSON keys, quoted in-image text, 1:1 structure, ' - ' not em-dashes, diacritics, output only translation. The summary generation process: 5 iterations; step 1 identifies 1-3 missing entities (relevant, ≤5 words, novel, faithful, anywhere; delimited by ';'); step 2 writes denser summary of identical length. Guidelines: first summary ~80 words, verbose, non-specific; rewrite for flow; make space via fusion/compression; remove uninformative phrases; self-contained; never drop entities; add fewer if needed. Output must be JSON: a list of 5 dictionaries with keys 'Missing_Entities' and 'Denser_Summary'." }, { "Missing_Entities": "no preamble; no fences; preserve structure 1:1; use exact same word count each summary", "Denser_Summary": "The French translation task preserves VERBATIM placeholders [X]/{x}, brand/model names, param flags (--ar,85mm), code/JSON keys, quoted in-image text, 1:1 structure, ' - ' not em-dashes, diacritics, output only translation (no preamble, no fences). The summary generation process: 5 iterations; step 1 identifies 1-3 missing entities (relevant, ≤5 words, novel, faithful, anywhere; delimited by ';'); step 2 writes denser summary of identical length. Guidelines: first summary ~80 words, verbose, non-specific; rewrite for flow; make space via fusion/compression; remove uninformative phrases; self-contained; never drop entities; add fewer if needed. Output JSON: list of 5 dictionaries with keys 'Missing_Entities' and 'Denser_Summary'; use exact same word count each summary." } ]