How this record is built
Evidence before labels.
Publication Radar follows selected AI sources. Bibliographic confirmation is separate from scientific validation and human editorial assessment.
One work, separate records
A preprint, accepted submission, and proceedings article retain their own identifiers, dates and evidence. Exact provider identifiers and conflict-checked DOI identities can reconcile records. A trusted explicit publisher link can connect manifestations. Similar titles or author names create review candidates; they do not merge works. Expanded journal articles can remain separate, linked works.
What dates mean
All papers use the earliest known public release. Preprint views use first posting. Conference views use proceedings publication, or an explicitly evidenced acceptance when publication is absent. Journal views use online publication. Recently indexed uses first observation by Pub-AI. Filters use the selected view’s date; partial dates use their earliest possible bound for sorting and filtering while retaining the supplied precision on screen. Future dates cannot lead a newly published feed.
Acceptance requires venue evidence
Appearing on OpenReview is not acceptance. We require an exact, verified public venue-status mapping and publicly readable fields. TMLR is a journal; conference presentation certifications do not turn it into a conference publication. An outage, missing page, or metadata edit never becomes a rejection, withdrawal, or retraction.
Versions and integrity notices
Only explicit source versions create revision events. Withdrawal and retraction facts are scoped to the affected record and require official target evidence. Integrity-notice coverage is limited: no comprehensive retraction monitor is claimed. Historical baseline events remain on timelines and do not enter the current change feed.
Human selections retain their scope
Publication Radar picks require an explicit human selection. World Models editor picks keep their original scope and explanations. Neither import automation nor an LLM can promote a paper. Metadata can appear without a generated summary; code links do not mean independent reproduction.
Coverage and reuse
Each partition has its own edition, track, check time and gaps. Initial imports can be partial. Abstracts are displayed only where reuse is established; otherwise the original record is linked. We do not mirror PDFs, collect reviews, infer anonymous authors, or display private fields. Thank you to arXiv for use of its open access interoperability.
For corrections, use the existing contact and submission route. Changes are reviewed and audited before publication.
Three distinct feeds
- New papers RSS · at most 100 items
- Publication changes RSS · at most 100 items
- Human editor picks RSS · at most 100 items