Instructions: You are given a factual claim and its associated structured data. Your…
- User
Instructions:
You are given a factual claim and its associated structured data. Your task is to rewrite the claim to make it clearer and more general, while still preserving its core factual meaning. Use the structured data to ensure accuracy.
Guidelines:
- Use the data provided — do not rely on external knowledge.
- Avoid evaluative or moral language (e.g., “better,” “worsened,” “concerning”).
- Keep the rewritten claim concise and factual.
- Ensure chronological references are accurate and properly ordered.
- If the claim or data references implausible or clearly erroneous content (e.g., years far in the future like 2059), return an empty string.
- Ensure the claim includes enough context to be understandable on its own.
- Avoid overly narrow or trivial claims unless necessary to retain factual meaning.
- If the original claim is confusing due to mixed-up data, fix it using the data provided.
Output Format:
Return the rewritten claim in JSON with the key "rewritten_claim". Use
jsonto wrap the JSON.{ "rewritten_claim": "..." }Input:
Claim: Russia's forest land carbon emissions have increased from 130th to 1st among OECD countries, rising from 11,981 thousand tonnes of CO2-equivalent in 2000 to 17,218 thousand tonnes of CO2-equivalent in 2019. Data: {'Erosion risk level': 'Not applicable', 'Measure': 'Forest land', 'Nutrients': 'Not applicable', 'Reference area': 'Russia', 'Unit multiplier': 'Thousands', 'Unit of measure': 'Tonnes of CO2-equivalent', 'Water type': 'Not applicable', 'highest_rank': 1.0, 'highest_value': 17218.478, 'highest_year': 2019.0, 'lowest_rank': 130.0, 'lowest_value': 11981.069, 'lowest_year': 2000.0, 'diff': 130.0}
- Assistant
{ "rewritten_claim": "Between 2000 and 2019, Russia's forest land carbon emissions increased from 11,981 thousand tonnes to 17,218 thousand tonnes of CO2-equivalent, moving from 130th to 1st rank among OECD countries." }
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.