> ## Documentation Index
> Fetch the complete documentation index at: https://docs.valyu.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Selected-paper experimental comparison

> Compare selected experimental papers in a cited evidence table and CSV, keeping device, method and measurement differences explicit, with Valyu DeepResearch.

**Example question:** What can we compare between these two perovskite-silicon tandem papers, and which measurements use different specimens or conditions?

**You get:** A paper-by-paper evidence table and CSV with device details, measurements, source locations and comparison limits.

**Start with:** Two to ten selected paper URLs or accessible files; mandatory inclusions; comparison fields; question to resolve.

This recipe starts with papers you have already selected. To discover a reading set first, use [finding and selecting papers](/use-cases/academic). For enzyme-assay condition extraction, see [assay method comparison](/use-cases/sciences/experimental-protocol-comparison).

## Run with DeepResearch

Set `VALYU_API_KEY` in your environment, then run this illustrative request. Change the question and CSV fields for your own task. The CSV fields below describe the requested output, not a saved result.

<Accordion title="Run the research request">
  ```bash cURL theme={null}
  curl --fail-with-body https://api.valyu.ai/v1/deepresearch/tasks \
    -H "x-api-key: $VALYU_API_KEY" \
    -H "Content-Type: application/json" \
    --data-binary @- <<'JSON'
  {
    "query": "Compare the two supplied original papers on perovskite-silicon tandem solar cells. Treat both as mandatory anchors. Extract architecture, specimen area, reported efficiency measurement and stability test evidence. Keep certified efficiency separate from the performance of the specimen used in stability tests. Distinguish component films, single-junction cells and complete tandems. Record missing conditions as not reported and inaccessible methods as not accessed. Do not rank devices from incompatible tests or infer details from review articles.",
    "mode": "standard",
    "research_strategy": "Use the supplied original papers and their supplements. Keep abstract-only access explicit. A failed full-text fetch is not proof that a measurement was never reported. Cite every row; mark missing or inaccessible evidence.",
    "report_format": "A concise evidence table, comparison limits and unresolved questions.",
    "deliverables": [
      {
        "type": "csv",
        "description": "One row per evidence item with the requested fields and supporting source URL.",
        "columns": [
          "paper_id",
          "device_or_specimen",
          "architecture",
          "measurement_method",
          "measurement",
          "unit",
          "test_conditions",
          "evidence_location",
          "source_url",
          "access_status",
          "comparison_limit"
        ]
      }
    ],
    "urls": [
      "https://www.science.org/doi/10.1126/science.adg0091",
      "https://www.science.org/doi/10.1126/science.adp1621"
    ]
  }
  JSON
  ```
</Accordion>

Save the returned `deepresearch_id`. Poll [Get status](/api-reference/endpoint/deepresearch-status) until the task completes, then read the report and download the CSV from `deliverables`.

## How it works

1. Record each paper identifier and whether full text and supplementary methods are accessible. Retain inaccessible papers in the selection list.

2. Extract device architecture, measurement method, specimen area and reported results separately for each paper. Attach section, figure or table locations when available.

3. Compare only compatible measurements. Label missing conditions and explain where differences in device or test design prevent direct comparison.

To build the workflow inside your own agent, use [Search](/search/quickstart) to find sources and [Contents](/guides/content-extraction) to read selected pages. Keep the CSV fields above as your extraction schema.

## Check the result

Check units, specimen identity, time units and the denominator of retained performance. Confirm all mandatory papers appear, including unreadable ones. This is selected-paper analysis, not an exhaustive systematic search.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.