video1.0<iframe src="https://www.loom.com/embed/6c4edb59ed714f1694d1bc65cca2d934" frameborder="0" width="1728" height="1296" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12961728Loomhttps://www.loom.com12961728https://cdn.loom.com/os/loom-media-content/loom/774aa224-135b-45a1-8cd6-bc58a0264ad1/6f868b40-e56c-4127-8486-7682bffe89c1/sessions-thumbnails-6c4edb59ed714f1694d1bc65cca2d934-101d894b5ac63c96.gif/binary?authToken=eyJraWQiOiJtaWNyb3MtZ3JvdXAvdGRwLW9zLWFyY2hldHlwZS9hOWI5NmI1OWlwM3JxOWRtIiwiYWxnIjoiUlMyNTYifQ.eyJicyI6Imxvb20tbWVkaWEtY29udGVudCIsImF1ZCI6InRkcC1vcy1hcmNoZXR5cGUiLCJuYmYiOjE3OTA5NDE3OTIsInRpIjoiMGE4Mjk0NTEtMWRmZi00MzAwLWI1MTYtNmVjOWIzYjFiZmRjIiwiYWRpIjoibG9vbS9sb29tIiwiaXNzIjoibWljcm9zLWdyb3VwL3RkcC1vcy1hcmNoZXR5cGUiLCJwaWQiOiI0OWU1NjExZi0xOWFhLTQ4MGQtOGNkZS1iYmEyYTdkNTg3ZjIiLCJ3YSI6ImZhbHNlIiwib2lkIjoic2Vzc2lvbnMtdGh1bWJuYWlscy02YzRlZGI1OWVkNzE0ZjE2OTRkMWJjNjVjY2EyZDkzNC0xMDFkODk0YjVhYzYzYzk2LmdpZiIsImV4cCI6MTc5MDk0MjM5MiwiaWF0IjoxNzkwOTQxNzkyLCJqdGkiOiI0YmYyM2U2Mi05MTYwLTQwNWQtYTkyZi1iYWE3ZDZkMjRhZjQifQ.sqI-ql36qa9CXjmUcjqBwGrmQyNZuoD-_8q_1IqARUrXgXYvBi1gr3xqwnzQ5epEYhrGeIenCktknhuqZEJXL8mYhn3VRmnt47qDzG-rLuQO9606Su65fPh3clwtEv9gbZ1ST4A6kcVm8eA-lDF8BYqv0fZFKbd4AWTsuHiiBNAZwaUWTIdqUNY19dmiKU-QbgkhD4cH6mGb00WpNHqjnrmQoB3jYnm6ckUCE9Z9kpZiO4Xc85bEZgdRUly3XR970DO0T06vqmOX583pXYy_7YS5GvKdkCdXwEVSTEhyTY8ILEy3P3XOKTFcY2CPyqQNSBMPbcfOh0jnywAxYFy1Sg&Expires=1790942392&Signature=Cj5gz9cjOrgZgupeMlDTGcsZxXJyPhhZFGBdLf-jARScnpdKs9AvsTD8mPiJbF6h01Kdp7Zgnb543jHJ7Whu05lHdX2bhgPizxzrELi9JozAL%7EaWZQT%7EYbapfLjib6mP-2Fjbbhj1LCdZh5sNU0FNJ8MOD0eLChtvP4WDgZfHNU-Kt6shEFpgMS9R0MGq7bX4j%7EPfQOSjaJ7SMb7zcMrlWi1P2jwrUAszF8VEVMlmijYYKbpOjpSLSpfrInrdEEhhfW0RgBWX0ZFQ18maFYvnNwR6c6dGquMWcig60EyCPWI5bs7xV51byyCoTpnmm8dTAPhsn6qRZQQbTE9Njsylw__&Key-Pair-Id=KQOSYIR44AIC0174.08This Loom presents “Investigation Certificates for Agentic Graph Rack” to solve exhaustive counting failures in standard Graph RAG. The author explains an orchestrator that routes questions by evidence completeness, bypassing similarity search for exhaustive counting and using a deterministic TigerGraph GSQL count to avoid unnecessary LLM reasoning. In benchmarks on 100 public questions, Plain Rack reached 25% accuracy with about 14,000 tokens per answer, Graph Rack improved to 43%, and their Hintik Pipeline achieved 92% accuracy while using about 42 LLM tokens per answer. They illustrate with a 2004 Summer Olympics counting example, showing that their certificate-bound approach provides mapped coverage and retrieved documents instead of guessing.