MetadataMatters!

The Good, the Bad, and the Repairable: What 127,166 TP53 Records Reveal About AI-Ready Scholarship

CAISc 2026 · Accepted conference paper

An AI-written review can sound authoritative without revealing whether the system saw a full paper, an abstract, or only a title. We ask how often bibliographic records provide the connections needed for grounded synthesis. We audited 127,166 TP53-related journal articles indexed by OpenAlex between 2000 and 2025 across five facets: provenance, people, organizations, funding, and access. Only 7.37% meet the strict criterion for all five. Citation count is moderately associated with the number of facets present (Spearman ρ = 0.42) but only weakly associated with passing all five (ρ = 0.131); among the 100 most-cited works, 7 pass. The result is not an artifact of the strictest cutoff: at 80% authorship-level ORCID/ROR coverage, completeness is 19.65% overall and 18.0% in the top 100. We connect these measurements to nine retrieval tasks, examine five highly cited records in an AI-authored TP53 review, and estimate what targeted metadata repair could achieve. People metadata is the largest bottleneck: completing it would make 36,125 additional records complete across all five facets. These are limits of one retrieval substrate, not of the underlying literature or every tool-enabled AI system.

Aadi Narayana Varma Dantuluri

Towards Nexus-Score: Metadata Gaps Limit Scholarly AI Attribution

arXiv · Version 1 · cs.DL

Artificial intelligence increasingly mediates how scientific work is discovered and credited. This study asks whether missing metadata prevents AI systems from attributing work correctly. In an initial boundary test, a system without access to the task-relevant paper list frequently returned identifiers outside that list, including fabricated ones. The authors then hid or restored author, institution, funder, reference, and text-access links in OpenAlex records while keeping the works and tasks fixed. Restoring the relevant link enabled the corresponding attribution, whereas restoring the wrong facet produced no correct answers across 469 completed mismatched tests. Missing connections led to invented answers, refusals, or exhausted tool budgets, and web search did not recover concealed author links. The findings motivate Nexus-Score as a record-level check that can expose repairable metadata gaps and prepare scholarly records for AI-mediated use.

Aadi Narayana Varma Dantuluri · Sushrut Thorat · Paras Chopra

aadivar/nexus-score v0.1.2: Typo-tolerant institution search

Zenodo · Version v0.1.2

Open-source software that turns Crossref metadata coverage into an actionable, 100-point view of how well publisher records connect research outputs, people, organizations, funders, and access. It has evaluated 27,000+ Crossref-registered publishers, while its current-era leaderboard focuses on metadata from the last two calendar years. David Worlock describes Nexus Score as a breakthrough in open-source scoring environments and highlights the surprising mix of high-quality metadata publishers near the top.

Provenance25 pts

References, update policies, similarity checks, and other signals supporting citation links and research integrity.

People20 pts

Author identifiers, especially ORCID iDs, and contributor roles.

Organizations15 pts

Affiliations and persistent organization identifiers such as ROR IDs.

Funding20 pts

Funder Registry identifiers and grant or award numbers.

Access20 pts

Licenses, full-text links, and abstracts supporting discovery and text mining.

Varma D Aadi Narayana

Using Technology to break cultural silos towards open data and open science

OpenResearch London · Francis Crick Research Institute

Presentation proposing that the cultural shift toward open data and open science must be continuous, supported by actionable feedback, education, shared infrastructure, and coordinated workflows. It imagines dynamic research records that keep procedures, data, null results, and reuse evidence connected, while recognizing researchers who improve and reuse them. In the founder’s evolving vision for Annexis, its focus on observable reuse is developing into Counter as a Service: a provenance aware way to understand how AI agents access and use scholarly records.

Aadi Narayana Varma D · Sheevendra Sharma

Metadata Pedigree

FORCE2017 · Berlin, Germany

Presentation and flash talk proposing a “DOI pedigree” that connects object level event data across the research lifecycle with researcher profiles. Its central vision is to recognize researchers for how they strengthen quality, reuse, reproducibility, curation, and repair, independently of how the underlying research is assessed. This separation between research assessment and researcher evaluation anticipates the founder’s wider vision for Annexis: infrastructure that makes otherwise invisible stewardship and reuse contributions visible and rewardable.

Aadi Narayana Varma D · Sheevendra Sharma

Transitioning “Open Data” from a NOUN to a VERB

PeerJ Preprints · Version 1

Preprint exploring how research assessment can be distinguished from researcher evaluation, and how that separation could change participation in the Open Data movement. It argues that lasting change needs shared infrastructure capable of aligning the priorities of researchers, institutions, funders, publishers, and service providers. That early proposition now informs the founder’s vision for Annexis: an actionable coordination layer that makes contributions, gaps, and opportunities for reuse visible across the scholarly record.

Aadinarayana Varma D · Sheevendra Sharma