Resolve people across documents

Documents often contain partial and inconsistent references to the same person: a bank statement, invoice, and payslip may use different formats, different name forms, and no shared customer identifier.

The same person across documents notebook uses the repository's sample PDFs to run the document path from file discovery to an inspectable resolution report.

What the workflow does

from arche import resolve_documents

report = resolve_documents("data/docs/*.pdf", jurisdiction="NG")
print(report.table())

The report masks values by default, retains extracted records and decisions, and shows the evidence behind each verdict. A high score can still be review when the evidence does not justify an automatic identity claim.

Run it locally

Install document support, then open the notebook from examples/notebooks so its relative paths resolve to the repository data.

pip install "arche-core[doc]"
cd examples/notebooks
uv run --package arche-core jupyter notebook 02_same_person_across_documents.ipynb

The sample PDFs are a demonstration, not a person-resolution benchmark. Review the extraction results before relying on a linkage decision. Resolution can only be as reliable as the identifiers and references extracted from the documents.

Use agents carefully

An agent may prepare document inputs, summarize a masked report, or present a review candidate to an authorised person. It must not turn review into a match, reveal data outside the approved boundary, or treat a document-derived decision as independent evidence without checking source provenance.

See the notebook release evidence for the execution check and limitations of this example.