Keep Your Receipts: Trustworthy Cleanup of Agent-Built Knowledge Graphs

Knowledge graphs are becoming the ground truth for agentic memory systems, and constructing them has become easy. Maintaining them has not: when a source is retracted, overruled, or erased, everything the pipeline built on top of it, entity merges, entailed edges, community summaries, must be re-examined, and a maintainer is caught between deleting too much and losing valid knowledge or deleting too little and serving contaminated content. Deployed pipelines take the timid extreme, removing only the directly extracted facts, which on our corpus reduces contaminated queries from 32.2\% to just 28.0\% under a real retraction and changes nothing under predicate revocation. Rebuilding is no repair either: two builds of the same 300-document corpus differ in 29,475 artifacts, five times the 6,248 that removing 61 retracted documents changes. We introduce PolyGraph: the constructor must emit receipts for every artifact during construction, alternative bundles of source evidence, each sufficient on its own, and a re-derivation audit double-checks those receipts by deleting the cited evidence and testing whether the constructor still produces the artifact. Cleanup then costs no model calls at all, milliseconds of set arithmetic over a 24,381-artifact graph. The audit shows the receipts are honest where their claims are narrow and checkable, 68\% and 85\% agreement at the merge layer, and fiction about two thirds of the time at the summary layer, almost always erring toward keeping dead content. Distrusting just that layer cuts contaminated queries to 0.8\% while retaining 99.8 to 100\% of the recall a perfect maintainer would return.