PROBLEM: A scientific model that spans many dimensions — materials, fluids, energy, biology — is built from many codes, datasets, and versions, and when the model changes, the record of what changed rarely keeps up, so results cannot be traced back to the model that produced them.
SOLUTION: A multidimensional model provenance and versioning record: every model version carries its code versions, dataset versions, parameter set, and machine, linked to every result it produced — so any result can be traced to the exact model that made it.
EVIDENCE BASIS: DOCUMENTED — Exascale modeling at ORNL and LLNL depends on named, versioned codes (BerkeleyGW, MFC) run on named machines (Frontier, El Capitan) with published system sizes and precisions; reproducibility in this field is defined by those records. DOCUMENTED — El Capitan's unified memory architecture and mixed-precision storage let over 100 trillion grid cells fit in memory, a configuration fact that must be recorded for any result to be reproduced. The combined multidimensional modeling system is an inventor concept only, UNVERIFIED until measured. Sources: ORNL; LLNL.
| Idea priority | Low |
| Needed By | Not sure -- Just thought it was cool |
By clicking the "Post Comment" or "Submit Idea" button, you are agreeing to the IBM Ideas Portal Terms of Use.
Do not place IBM confidential, company confidential, or personal information into any field.