Prototype and reproduction package for "Organizing Contestable Evidence in AI-Mediated Archives: A Provenance-Aware GraphRAG Framework for Knowledge Organization"

This package accompanies the article "Organizing Contestable Evidence in AI-Mediated Archives: A Provenance-Aware GraphRAG Framework for Knowledge Organization", published in Knowledge Organization. It contains a reference implementation of the evidence, provenance, and causal graphs proposed in the article, the cross-layer validation rules (R1 to R3), the human-confirmation rule (H.1 and H.2), dependency-adjusted source clustering, the contestable evidence bundle, bundle-level checks (B.1 to B.4), and the scoring rules.

The package includes a synthetic test case with simulated agents, a suite of 36 tests covering every named check and cases that must pass, and the encoding of one authentic archival record, a 1966 approval held by the Shanghai Municipal Archives (A038-002-00759-0030), taken from the authors' research notes. All outputs reported in the article are included, together with phase timings for replicated graphs.

The code requires Python 3.9 or later and uses only the standard library. Run "python3 run_poc.py" to reproduce the outputs, "python3 tests.py" to run the tests, and "python3 case_archival_item.py" for the archival example. The README describes each file, the check and handling codes, the relation between the synthetic case and real archival environments, and the scope of the timing results.