sec-edgar-pipeline

SEC EDGAR Pipeline

Overview

This pipeline is centered on edgar-analyzer and the EDGAR data sources. The core loop is: configure credentials, create a project with examples, analyze patterns, generate code, run extraction, and export reports.

Setup (Keys + User Agent)

Use the setup wizard to configure required keys:

python -m edgar_analyzer setup
# or
edgar-analyzer setup

Required entries:

• OPENROUTER_API_KEY
• (Optional) JINA_API_KEY
• EDGAR user agent string ("Name email@example.com")

End-to-End CLI Workflow

# 1. Create project
edgar-analyzer project create my_project --template minimal

# 2. Add examples + project.yaml
# projects/my_project/examples/*.json

# 3. Analyze examples
edgar-analyzer analyze-project projects/my_project

# 4. Generate extraction code
edgar-analyzer generate-code projects/my_project

# 5. Run extraction
edgar-analyzer run-extraction projects/my_project --output-format csv

Outputs land in projects/<name>/output/.

EDGAR-Specific Conventions

• CIK values are 10-digit, zero-padded (e.g., 0000320193).
• Rate limit: SEC API allows 10 requests/sec. Scripts use ~0.11s delays.
• User agent is mandatory; include name + email.

Scripted Example (Apple DEF 14A)

edgar/scripts/fetch_apple_def14a.py shows the direct flow:

1. Fetch latest DEF 14A metadata
2. Download HTML
3. Parse Summary Compensation Table (SCT)
4. Save raw HTML + extracted JSON + ground truth

Recipe-Driven Extraction

edgar/recipes/sct_extraction/config.yaml defines a multi-step pipeline:

• Fetch DEF 14A filings by company list
• Extract SCT tables with SCTAdapter
• Validate with sct_validator
• Write results to output/sct

Report Generation

edgar/scripts/create_csv_reports.py converts JSON results into:

• executive_compensation_<timestamp>.csv
• top_25_executives_<timestamp>.csv
• company_summary_<timestamp>.csv

Troubleshooting

• No filings found: confirm CIK formatting and filing type (DEF 14A vs DEF 14A/A).
• API errors: slow down requests and confirm user-agent is set.
• Extraction errors: regenerate code or use manual ground truth in POC scripts.

Related Skills

• universal/data/reporting-pipelines
• toolchains/python/testing/pytest