citation-drift-lab Tool Detects When Cited Sources Change After an Answer Is Given

A developer has released citation-drift-lab, a small offline tool designed to expose how answers built on document citations can become unreliable when source documents are later modified. The tool ingests Markdown files, records exact character offsets and SHA-256 hashes at the time of answering, then audits those citations against updated versions of the same source. A built-in demo illustrates the problem using two versions of a policy document, where a project limit changes from 30 to 20, an exception clause is deleted, and a section is relocated. The audit identifies which claims are materially affected and flags them accordingly, while questions lacking sufficient evidence are recorded as abstained rather than answered incorrectly. The project requires no API key, external model, or network connection, making the drift-detection process fully reproducible in a local environment.