Digital-Mailroom: appendix
Public corpus · partner boundary
Supporting notes for the Digital-Mailroom case study. The capstone publishes no measured results here; these tables describe the data boundary and where the public corpus comes from.
Public and non-public
| Public on this site | Not published |
|---|---|
| The capstone problem framing and pipeline description | The partner’s documents |
| The public-source corpus and its provenance | Internal partner metrics |
| The classifier work that feeds the pipeline | Non-public evaluation results |
| Reproduction notes for the public pieces | Anything under a partner confidentiality agreement |
The industry partner scoped the problem and met with me through the capstone. Their data was never used to build the corpus.
Corpus sources
The public corpus has 3,302 documents. No partner documents are included.
| Class | Rows | Source | Terms |
|---|---|---|---|
| Insurance claim | 1,100 | CMS DE-SynPUF, GNOTHEIA, BDR, INSURBIAS | Apache-2.0, MIT, CC BY 4.0 |
| Correspondence | 1,000 | Deduplicated Enron email | Research use; real named-person PII |
| Contract | 600 | CUAD v1 (509) and SEC EDGAR EX-10 (91) | CC BY 4.0; public domain |
| Corporate record | 450 | SEC EDGAR S-1 and 8-K exhibits | Public domain |
| Merger agreement | 152 | MAUD v1 | CC BY 4.0 |
The Corpus EDA appendix records label provenance, configuration counts and context bands. The dataset card carries the versioned citation.
Where the code lives
The linked repository is the integration workspace: one lockfile and one virtual environment wiring the standalone packages together. Package code is authored and released from its own repository, then imported, so a fix belongs upstream rather than in the workspace copy.
Status
Measured capstone results remain unpublished while the project is running, with expected completion in December 2026. This appendix will not carry numbers before the partner-facing report exists.