Versioned, checksummed datasets used by labagent lab packs. Files are published as release assets, not committed. labagent pins each release by SHA-256.
Section-split text from recent SEC 10-K annual reports, chunked for retrieval (RAG) workshops.
| Passages | 1,880 |
| Filings | 10 (AAPL, CVX, JPM, PFE, WMT; two fiscal years each) |
| Sections | Item 1 Business, Item 1A Risk Factors, Item 7 MD&A |
| As of | 2026-10-22 |
| Size | about 0.5 MB compressed |
Files (in the release): sec-10k-2026.10.chunks.jsonl.gz, sec-10k-2026.10.manifest.json, SHA256SUMS.
Record format (one JSON object per line):
id, title, content, meta with ticker, company, form, fiscal_year (year of the period end date), item, section, filed, url (the filing on sec.gov).
Use with labagent (the rag-sec-filings pack ships in labagent 0.2.0, not yet on PyPI)
pip install 'labagent[rag]'
labagent launch rag-sec-filings --name sec --param profile=<databricks-profile>Verify a download
shasum -a 256 -c SHA256SUMS- JPM and CVX have no Item 7 (MD&A). Those filings place or incorporate the discussion elsewhere, and the parser only reads Item 7.
- Tables are flattened to text. Do not rely on numbers that come from tables.
- A few chunks are bare headings or page numbers, and a few exceed the 1,200 character target.
- Walmart's fiscal year ends in January, so its
fiscal_year2026 is the year ended January 2026. - No embeddings are included: embed with the model of your choice.
Text is derived from filings published on the SEC's EDGAR system. Each record links to its original filing. Please credit the SEC as the source. Provided as is, with no warranty. Informational only, not investment advice.