← Dataset guides

Python 3.10+ or n8n · Updated October 10, 2026

Export DeepSearchQA to CSV for an agent evaluation

Preview the DeepSearchQA schema, retrieve benchmark rows with your DataBazaar account, and export problem, category, answer and answer type to CSV.

Using n8n? Import the ready-to-run workflow.

1. Check the source and sample

Open DeepSearchQA to inspect its current availability, schema, documentation and sample assessment. Source: publisher dataset card. Declared license: Apache-2.0.

Expected fields: problem, problem_category, answer, answer_type.

Listing availability and source files can change. The commands fail visibly if the expected fields or access are unavailable.

2. Run a public sample check

Download the Python script and inspect the small sample without signing in.

curl -fsS --max-time 20 https://databazaar.io/examples/dataset-workflow.py -o dataset-workflow.py
python3 dataset-workflow.py deepsearchqa

The output says scope: sample_only. It reports sample rows and columns; it neither downloads nor certifies the full dataset.

3. Sign in and export real rows

Create an account, then create an API key. Store it in DATABAZAAR_API_KEY in your local environment. Keep credentials out of notebooks and shared files.

python3 dataset-workflow.py deepsearchqa --authenticated --limit 1000 --output deepsearchqa.csv

A CSV with the four benchmark fields and a JSON summary of the categories in the returned rows. The listing declares 900 records; the script requests at most 1,000 and reports the actual count. A successful query does not independently establish that the listing is complete.

This is an authenticated query. The script never buys data and refuses to overwrite an existing output file. HTTP 401 means account authentication is needed; other access errors require checking the current listing and permissions.

4. Use the output carefully

Pass only the problem field to your agent. Store its answer separately and use the reference answer only when scoring. Follow the source benchmark’s evaluation protocol; exact string matching is not an equivalent evaluator for multi-answer research questions. This guide prepares the data, it does not run or score a model.

The public sample is not representative of every category. Do not train on the evaluation set or leak reference answers into the agent prompt. Record the retrieval date, returned row count and response fingerprint so later comparisons use the same input.

How sample quality is assessed explains what the score can and cannot establish.

Export with n8n

This independent DataBazaar example uses built-in n8n nodes to produce the same four benchmark fields as a CSV. It prepares evaluation inputs; it does not run a model. Full retrieval requires a DataBazaar account, even though this dataset is free.

  1. Download the workflow JSON and use Import from File in a new n8n workflow.
  2. Create an account and API key.
  3. Open Retrieve benchmark rows. Select Generic Credential Type → Header Auth and create a credential with name Authorization and value Bearer YOUR_DATABAZAAR_API_KEY. Keep the key in the credential, out of shared workflow files.
  4. Choose Execute workflow, then open Create CSV file and download its data binary output as deepsearchqa.csv.

Tested with n8n 2.42.6 on October 10, 2026: all 900 rows and four columns matched Google’s source CSV, including its literal None answers. The workflow stops on an unexpected schema, row-count mismatch or a listing that is no longer free. It is specific to this benchmark; changing only the dataset ID is not enough to use it with other listings.

Read the workflow source, troubleshooting and verification. No purchase or model call is made. Your n8n hosting has its own costs.

Inspect this dataset →Connect your agent →