We model the contribution of citation cij from paper i to paper j as a Bernoulli draw whose success rate πj reflects the latent reliability of j. Under the prior that{' '} down-weighting works as a Bayes prior on citation utility, with the replication rate τj as a sufficient statistic for πj, we obtain the closed-form posterior in Equation (4).
Empirically, this re-weighting redistributes ≈18.6% of cumulative citation mass away from the long tail of un-replicated, low-tau papers. Section 4 evaluates the metric on a held-out set of 14,000 neuroscience papers spanning 2010–2023.
Edge cases.{' '} Two pathologies require explicit handling: (i) self-citation chains within a single lab, and (ii) citation rings that arise from venue-specific norms. We treat both as structured covariates rather than removing them outright.
{/* Margin pin */}
{`@article{lin2024rwif,
author = {Lin, C. and Harriot, M.},
title = {A reproducibility-weighted citation index for neuroimaging},
journal = {Cell Reports Methods},
volume = {5}, number = {11}, pages = {100918},
year = {2024}, doi = {10.1016/j.crmeth.2024.0918}
}
@article{yamamoto2025sleep,
author = {Yamamoto, K. and Tanaka, H. and Mori, J.},
title = {Sleep depth from astrocyte calcium dynamics --- RIKEN replication},
journal = {Nature Neuroscience}, volume = {28}, pages = {411--423},
year = {2025}, doi = {10.1038/s41593-025-01882-1}
}
@misc{harriot2025astro,
author = {Harriot, M. and Singh, L. and Okafor, R. and others},
title = {Astrocytic gating of slow-wave sleep across cortical depth},
howpublished = {UB-Chain dPID·a92f.04}, year = {2025}
}
@misc{patel2025sparse,
author = {Patel, A. and Mehta, R.},
title = {Sparse decoding of cortical dynamics with scaled-attention transformers},
eprint = {2503.18112}, archivePrefix = {arXiv}, year = {2025}
}`}