The recent inquiry about NeurIPSanonymous repository etiquette resonates with earlier debates such as [R] ICML Anonymized git repos for rebuttal(/post/r-icml-anonymized-git-repos-for-rebuttal-cmnms7gud0vvfcf93uuq8fawk) and [NeurIPS openreview - can I upload paper pdf after abstract deadline or should I upload something first to be able to update it later? [D]](/post/neurips-openreview-can-i-upload-paper-pdf-after-abstract-dea-cmoskv9x109jljfqb3viyt331) as well as [Submitting to top ML Conferences without Sharing code [D]](/post/submitting-to-top-ml-conferences-without-sharing-code-d-cmogsh23t13udzxsx1ehm3byw). The question centers on whether a post‑deadline clarification in an anonymous code README that merely corrects a label can be considered a permissible annotation or an impermissible revision of the submitted manuscript. The submitter notes that the numerical results remain unchanged, but the corrected label alters the interpretive narrative of the reported metrics. This distinction matters because NeurIPS policies draw a bright line around any post‑deadline modifications that could affect the paper’s scholarly record, even when the underlying data stay the same.
Why does this nuance command attention from the broader community? First, the integrity of the anonymous submission pipeline hinges on the ability of reviewers to evaluate work without being influenced by post‑hoc narrative tweaks that might shift the perceived contribution. If authors can quietly adjust interpretive language in a README after the review window closes, they gain a subtle avenue to shape how their results are framed, potentially tipping the balance in borderline cases. Second, the practice of maintaining a fully anonymous repository while still offering a public correction creates a tension between transparency and the strict anonymity guarantees that reviewers rely upon. The community has previously explored similar tensions in the context of code sharing and rebuttal materials, as illustrated by the linked discussions, and those conversations underscore the need for clear, pre‑approved channels for any clarification that could affect evaluation.
What guidance can be offered to authors who find themselves in this spot? The safest route is to treat any post‑deadline adjustment as a communication rather than a revision of the paper itself. A brief note in the README that reads “labels corrected; numbers unchanged” can be useful, but it should be framed as an informational addendum that does not alter the manuscript’s submitted version. Submitting a separate comment on the OpenReview thread to flag the correction can provide a transparent audit trail without jeopardizing the anonymity of the submission. Moreover, if the authors anticipate the need for such clarifications, they might consider embedding the corrective note directly in the supplementary material before the deadline, thereby avoiding any perception of retroactive change. In all cases, the emphasis should remain on preserving the original submission’s integrity while still satisfying the community’s demand for accurate reproducibility.
Looking ahead, the tension between anonymity, reproducibility, and post‑