Turning Evidence into Verified Rule Graphs

video1.0<iframe src="https://www.loom.com/embed/2a931a78ccb54256880b9869407b93f0" frameborder="0" width="1920" height="1440" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>14401920Loomhttps://www.loom.com14401920https://cdn.loom.com/os/loom-media-content/loom/9564676e-9cec-480f-9105-84934ac56acb/c304714c-b971-447d-9368-f90935cd6f75/sessions-thumbnails-2a931a78ccb54256880b9869407b93f0-c90f7a05d19d5b20.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS9xcGE2cG03NzRjbzIzZ2w1IiwiYWxnIjoiUlMyNTYifQ.eyJicyI6Imxvb20tbWVkaWEtY29udGVudCIsImF1ZCI6InRkcC1vcy1hcmNoZXR5cGUiLCJuYmYiOjE3OTQyNTU4OTEsInRpIjoiZjgxZmEzMjctMjZkYy00MTc5LThmMDgtYTFjNGE3NmI3ZTJlIiwiYWRpIjoibG9vbS9sb29tIiwiaXNzIjoibWljcm9zLWdyb3VwL3RkcC1vcy1hcmNoZXR5cGUiLCJwaWQiOiJlYjA3OTVmOC03M2QzLTQ0Y2MtYjAxYi0xMmExYjc0MDI0NjciLCJ3YSI6ImZhbHNlIiwib2lkIjoic2Vzc2lvbnMtdGh1bWJuYWlscy0yYTkzMWE3OGNjYjU0MjU2ODgwYjk4Njk0MDdiOTNmMC1jOTBmN2EwNWQxOWQ1YjIwLmdpZiIsImV4cCI6MTc5NDI1NjQ5MSwiaWF0IjoxNzk0MjU1ODkxLCJqdGkiOiIzZTIwN2I5MS02OTQzLTRmMjktYmZlYS03MTQ5ZjMyMDUwN2IifQ.tdpS8RWtUFWxbu3gio2MyqovpL1B9zGKNP4sndaiJz0ri_S71FCWJhkjEPE0lCBs9aWTD0UUxiUESR61iHLIdhgMQi80vw8NqTS2HcVKsUU7NZss1Ep9LccNV8jbcsjhr5DBJVhY4ce3_LzPdHY_yX52zJ4HlkFAWam5fEG-FTHA3u9JI259SYOLilUOWHt14BAcmsrXYT6gbM0EXhWWgOwiqyPMhrI2UKBa80QBTWXBRmcBHePyFvd7jgsZWW2J835wnkXhNL6MMy88Zy6GwRMhrjgw1giXAbRU4WJChI8B35BzlSGKkdRWwlS6zDA3rXxjjyTvfa0o-L32NiksqA&Expires=1794256492&Signature=NuZOxf-1XAUA99QV3rPQVywnEPbZ1ZFskRT5%7E50uqxEuOGqzM-13OeVbjDJcPCcE502tp1ogA-BuE%7EVBsJOVrIkwKgYnevURHjeH-WoaXHVtA-rlFmDjsMfjb-jH1yObCnmfxYbtud5cz3BYXh3QmVly812W%7EU2XFVJpKBJGULZGDIcAkKE9sm3GHgnZwyGoYcEpFvNGKfc3hlXtyJG9lgzftpo8er8SYeVOp9TGhitVF4dzAPueGBG0hOI4NyKpdRIx82novXszoY34o3Hy2sdLEZdsE61yheBBXRwUWjN3mps5yMsxasYjWNf1WBTqx2tBnIJ-HTxLVtE0I3gR0A__&Key-Pair-Id=KQOSYIR44AIC0154.2Turning Evidence into Verified Rule Graphs - ProofGraphThis Loom explains how an AI system extracts and verifies legal and contract terms from scanned documents using a graph-based workflow. The speaker notes that over 2000 code decisions have involved AI-hallucinated content, including a case where forms were sanctioned $31,000 over a single unchecked handoff. They describe how Muse Spark reads scan lines into clause-level rules pinned to exact routes, then runs a second pass to catch what was missed by aligning code back to the page and building a typed graph. Verification requires both passes to be empty when something is not found, and it labels each claim and rule before deciding based on the weakest link.