Skill: discourse-consistent document translation

[Case 2]

• Initial Environment: A document-level machine-translation benchmark evaluates whether prompts and model variants preserve entity consistency, referential expressions, coherence, and phenomena such as deixis, ellipsis, lexical cohesion, and zero pronouns.
• Real Question: "Translate the document while preserving discourse phenomena across sentences."
• Why this demonstrates the capability: The benchmark is explicitly built to expose failures that look acceptable at sentence level but become wrong at document level. A model must use previous context to resolve omitted subjects, maintain lexical consistency, and avoid discourse-breaking pronoun choices. This is exactly the capability a document agent needs for serious translation work.

Pipeline Execution Instructions

To synthesize data for this capability, you must strictly follow a 3-phase pipeline. Do not hallucinate steps. Read the corresponding reference file for each phase sequentially:

1. Phase 1: Environment Exploration Read the exploration guidelines to discover raw knowledge seeds: references/EXPLORATION.md

2. Phase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory: references/SELECTION.md

3. Phase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data: references/SYNTHESIS.md