Open Dataset: AI-Analyzed SEC Filings with 30-Day Outcomes
StockHuntr publishes a free, open dataset of SEC filings analyzed by AI — one row per 10-K, 10-Q, or 8-K, each carrying an AI concern score, a management-sentiment score, and (where the 30-day window has elapsed) the realized 30-day market-relative return. That pairing of AI-derived features with a realized market label makes it a ready supervised-learning dataset for studying how filing content relates to subsequent returns. Released under CC-BY-4.0. Snapshot as of September 25, 2026 (generated September 28, 2026).
At a glance
| Total analyzed filings | 17,549 |
| With AI concern/sentiment features | 17,549 |
| With a realized 30-day outcome (label) | 4,009 |
| With a persisted live prediction | 3,918 |
| Distinct companies | 710 |
| Filing date range | 2023-12-13 → 2026-09-25 |
| Form type | Count |
|---|---|
| 8-K | 12,987 |
| 10-Q | 3,467 |
| 10-K | 1,094 |
Columns
Both files carry the same columns; JSONL is one JSON object per line.
| accession_number | SEC EDGAR accession number (unique filing id) |
| ticker | Stock ticker |
| company_name | Company name |
| sector, industry | GICS-style classification (may be null) |
| form_type | 10-K, 10-Q, or 8-K |
| filing_date | Date filed with the SEC (YYYY-MM-DD) |
| report_date | Period end date, if applicable |
| filing_url | Direct link to the filing on SEC EDGAR |
| predicted_30d_alpha | Model's predicted 30-day return minus S&P 500 (%) |
| predicted_30d_return | Model's predicted 30-day raw return (%) |
| prediction_confidence | Model confidence, 0–1 |
| concern_level | AI multi-factor concern score, 0–10 (higher = more concerning) |
| sentiment_score | AI sentiment of management discussion, -1 to +1 |
| eps_surprise_pct | Reported EPS vs consensus (%), where available |
| revenue_surprise_pct | Reported revenue vs consensus (%), where available |
| actual_30d_alpha | Realized 30-day return minus S&P 500 (%), if the window has elapsed |
| actual_30d_return | Realized 30-day raw return (%), if the window has elapsed |
| outcome_known | 1 if the realized outcome is present, else 0 |
Live data via the MCP server
For programmatic and agent access to live data, StockHuntr runs a remote Model Context Protocol (MCP) server. Point any MCP-aware client (Claude, ChatGPT, or your own agent) at:
https://www.stockhuntr.net/api/mcp
| get_latest_filings | Recent filings, filterable by ticker and form type. |
| get_filing_analysis | Full AI analysis for a single filing by accession number. |
| get_company | Company snapshot: sector, fundamentals, recent filings. |
| search_companies | Find tracked companies by name or ticker. |
| get_top_signals | The model's highest-conviction 30-day prediction signals. |
| get_model_track_record | Predicted vs realized outcomes and accuracy. |
Methodology & sources
Predictions come from a Ridge-regression mixture-of-experts model (price momentum, analyst activity, AI concern/sentiment, EPS surprise, filing type, tone shift vs. the prior filing, and macro regime). Full methodology and accuracy figures are on the FAQ & methodology page. Filings come from SEC EDGAR (public-domain U.S. government data); market data from Yahoo Finance; macro data from FRED.
How to cite
StockHuntr (2026). AI-Analyzed SEC Filings with 30-Day Predictions & Outcomes. CC-BY-4.0. https://www.stockhuntr.net/dataset
For research and education only. Nothing here is investment advice. Predictions are model outputs with known error; realized outcomes are included precisely so the model can be evaluated honestly. Past performance does not guarantee future results.