Bulk delivery and file formats
Download a licensed quarterly package and verify what you received.
For a complete walkthrough, create a key and verify your first download.
Discover the package
curl --fail-with-body --silent --show-error \
'https://sloedata.com/api/data-feed' \
-H "Authorization: Bearer $SLOE_API_KEY"The manifest identifies the package period, files, row counts, byte counts and SHA-256 hashes. It supplies authenticated download paths and an archive path. Request a specific published quarter with period=YYYY-QN; do not assume the latest website report also exists as a full bulk package.
Download an archive or a file
curl --fail-with-body --location \
'https://sloedata.com/api/data-feed?file=archive' \
-H "Authorization: Bearer $SLOE_API_KEY" \
--output sloe-quarterly-data.tar.gzFor one table, use a filename actually listed in the manifest, such as file=institutions.csv. Verify downloaded bytes and hashes against that manifest before importing. A broker redirect does not itself prove the file downloaded successfully.
The broker issues a short-lived token tied to the requesting customer's key. Redeem it with the same bearer key. Use the returned expiry and fetch a new link if needed. Replay protection depends on the configured server-side store; expiry and customer-key binding remain enforced. Treat the broker URL as temporary sensitive material. Do not store it as a permanent data URL or forward it to another customer. Inspect redirects before sending credentials to any new host; do not use --location-trusted to forward a key indiscriminately.
Available formats
| Delivery | Current format |
|---|---|
| HTTP reads | JSON |
| Quarterly tables | CSV; institution records also available as JSONL |
| Full package | tar.gz, including schema, source-period and license files |
| Native Parquet export | Snappy Parquet files listed in each published manifest |
Published native Parquet delivery
Native Parquet delivery is live with API 1.3.0, verified November 4, 2026. It uses the existing bearer-authenticated broker and private file-delivery flow. Once a package manifest lists a .parquet file, request that exact filename with the same key. The published manifest is the authority for the exact files and periods available.
The published delivery includes Snappy-compressed Parquet counterparts for historical Q1/Q2 tables and a separately scoped Q3 financial cohort: 4,154 institutions and 38 current metrics. That Q3 cohort is not a complete final-quarter census. The default complete historical bulk package remains Q2 until its own release changes.
The files preserve string identifiers, null-versus-zero distinctions and exact financial metrics using DECIMAL(38,12). Mixed-unit original account values retain their source text. Use each file's schema and source/release metadata when reading it.
For a package that lists institutions.parquet, the request shape is:
# Select a published period and a file returned by its manifest.
curl --fail-with-body --location \
'https://sloedata.com/api/data-feed?period=YYYY-QN&file=institutions.parquet' \
-H "Authorization: Bearer $SLOE_API_KEY" \
--output institutions.parquetReplace YYYY-QN with an available period from the manifest. Reuse the same short-lived broker precautions as other downloads. Read release status before depending on a newly added format.
Converting an existing CSV download
Prefer the native Parquet file when listed in your manifest. Conversion of another licensed CSV remains a downstream step. After a licensed download, your ETL can convert a table using an explicit schema. This illustrative example requires pandas and pyarrow; conversion does not grant extra storage or redistribution rights.
import pandas as pd
# Set identifier and value types to the downloaded schema before production use.
frame = pd.read_csv('institutions.csv', dtype={'charter': 'string'})
frame.to_parquet('institutions.parquet', engine='pyarrow', index=False)Check the actual identifier column in schema.json; the code above illustrates a table with a charter column. Explicitly type all identifiers and financial decimals for your real table. Compare row counts, nulls and values after conversion, and retain the source manifest with the Parquet object.
For higher-volume access and warehouse use, confirm your Data Access agreement. MCP interactive research access does not include bulk exports or warehouse synchronization.