Parquet export

Columns

The file has a fixed set of columns: id, ref, status, final_decision, accept, reject, rating_n, rating_mean, result, result_agreement and data. result is the answer as text, result_agreement is a number between 0 and 1 (0 when the question has none), and data holds your own columns as one JSON string, so every campaign exports with the same schema.

Reading it

import json
import pandas as pd

df = pd.read_parquet("campaign.parquet")
df["text"] = df["data"].map(lambda d: json.loads(d)["text"])
df[["text", "result", "result_agreement"]].head()

When to choose it

Pick Parquet for large campaigns and analysis. Pick JSONL when the output goes straight into a training script, and CSV when a person will open it in a spreadsheet.