Disclosure structure, not text intensity: What text-based greenwashing proxies capture in U.S. public-firm annual reports
收藏资源简介：
# Final results replication package This archive reproduces the results reported in the final manuscript and supplementary appendix from archived derived data, and it also includes a full public-source rebuild workflow. ## Package goals 1. **Exact archived-results replication**: rebuild every manuscript and appendix table/figure from the archived derived inputs included in this package. 2. **Public-source end-to-end rebuild**: reconstruct the pipeline from public SEC and EPA sources plus model access, without redistributing downloaded source files. ## What is included - End-to-end pipeline code (`semantic_inflation/`) - Minimal configuration and assets needed to rerun the workflow - Sanitized archived derived inputs used to regenerate the manuscript and appendix exhibits - Scripts to rebuild the reported exhibits and verify the rebuilt outputs - Public filing identifiers and metadata needed to reconstruct the SEC input universe without redistributing downloaded filings - Sanitized reference copies of the manuscript and supplementary appendix used to define the target exhibits ## What is intentionally not included - Downloaded SEC filing HTML files - Downloaded EPA bulk files - Local file-system paths from the development environment - Author names or manuscript title-page metadata - Raw machine-readable request/response logs containing filing text excerpts ## Quick start: exact archived-results replication From the package root: ```bash uv run python scripts/rebuild_results.py --output reproduced_results uv run python scripts/verify_results.py --expected expected_results --actual reproduced_results ``` This path does **not** call external APIs and does not require downloaded source files. ## Public-source end-to-end rebuild The package also includes a full rebuild script: ```bash export SEC_USER_AGENT="Researcher Name researcher@example.org" export OPENAI_API_KEY="..." bash scripts/run_public_rebuild.sh ``` This workflow downloads: - SEC annual report filings indexed by the pipeline - EPA GHGRP bulk data - EPA ECHO case and FRS bulk downloads The public-source rebuild reconstructs the full workflow, including contextual coding. Exact numeric agreement with the archived results may depend on continued availability of the archived model endpoint and provider behavior. The archived-results path above is the exact replication route for the published tables and figures. ## Main outputs - `expected_results/`: archived reference results included in the package - `reproduced_results/`: generated when you rerun the archived-results workflow - `public_rebuild_results/`: generated when you rerun the public-source workflow - `reference/`: sanitized manuscript and appendix reference files ## Reproducibility notes - All file paths in this package are relative. - Tables are stored as editable CSV files. - Figures are regenerated from source CSV files and exported as SVG, PDF, and PNG. - `docs/exhibit_map.csv` links each manuscript exhibit to its exact source file and output file. - `docs/inline_results_crosswalk.csv` maps key in-text numerical claims to the exact exhibit and data field that support them. - `docs/checksums.sha256` records hashes for archived inputs, scripts, and final exhibit files. - `docs/code_scope.md` describes which modules are required for exact archived replication and for the public-source rebuild workflow.
