Do funder records lead to checkable findings? A data-feasibility study of AI awards at UKRI, NSF and NSERC
Funder-grounded question answering requires an award record, a traceable association with a publication, and a finding supported by that publication's text. We compare the availability of this evidence across UKRI, NSF and NSERC, using outcome-independent, classification-defined samples of 100 awards per funder. At least one candidate chain was available for 68 UKRI awards (Wilson 95% interval 58.3–76.3%), 51 NSF awards (41.3–60.6%) and one NSERC award (0.2–5.5%). UKRI provides identified publication associations; NSF provides PI-reported lists with weaker identifiers; NSERC provides no publication-list channel. Under the provisional Canadian acknowledgement-and-recipient rule, only three award–publication pairs were confirmed, while 43 of 52 discovered pairs lacked an accessible acknowledgement. Seeded agent readings found genuine quantitative findings in 32/40 UKRI and 30/40 NSF candidates; these are not human verification. NSERC is reported as infeasible for the planned funder-grounded QA study under the current rule and collection regime. Author gate checks remain pending. This is a data-feasibility study, not the planned cross-funder QA experiment. Domain, award-category and closure rules necessarily differ, preventing a causal comparison of national research performance. The results locate infrastructure-dependent evidence gaps; an award–publication association never establishes that the award financed a particular finding.
Authors
- S. M. Abdulla Al Mamun
Institutions
- University of Manchester (GB)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24369147
- Primary Topic
- scientometrics and bibliometrics research
- Type
- preprint