CAVE-Onc: Graph-constrained agentic validation for cross-domain contradictions in CDISC oncology submissions
Jaime Yan · 2026
Read CAVE-Onc: Graph-constrained agentic validation for cross-domain contradictions in CDISC oncology submissions by Jaime Yan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/plos-pone-0350376.txt.
Description
Jaime Yan presents a validation system for catching logical contradictions inside the data packages that drug companies submit to regulators, in this 2026 paper. The industry's standard checker verifies each data domain separately, so a cancer response rating that conflicts with the underlying lesion measurements sails through. Yan's engine rebuilds the submission as a knowledge graph. It applies a layer of declarative rules, including cross-domain ones modeled on oncology response criteria. An AI agent layer handles the trickiest verification table.
On clean data the new layer flagged almost wholly different material than the checker, showing the two are complementary. On data seeded with twenty contradiction types, it caught every one, while two industry validators caught only the structural cases and none of the cross-domain response contradictions. Three experts endorsed the archetypes. Yan is careful to call this a demonstration of expressiveness, not a real-world detection rate.
Catalog record
| CREATOR | Jaime Yan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50420 |
| LICENSE | CC_BY |
| RIGHTS | CC BY |
| SOURCE | PLOS |