video1.0<iframe src="https://www.loom.com/embed/92f2d2adabc1415eaccbb111f645c8bf" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/92f2d2adabc1415eaccbb111f645c8bf-a3f2a62f3a6a044a.gif98.395This Loom demonstrates how an in-graph agent that remembers can be manipulated through seeded false claims and how contamination is traced. The attacker plants the claim that ratings are unreliable for low-vote movies and finds that top three movies of 1999 are unchanged, so the lie is reported and trust increases without anything looking wrong. A second test counts movies where the filter changes results, causing quarantine immediately and tracing a red contamination tree; the child trust is halved and returns to conventional. The presenter concludes that NGRAV traces which memory is infected and shows how contamination is contained.