Search uses a compact index of words and their positions in titles and abstracts. Only the index files needed for your keywords are fetched; full paper details load one results page at a time. Unicode and punctuation are normalized; simple plural endings are folded. Comma-separated phrases must all occur as contiguous whole-word phrases. The Python app uses the same membership rule; this demo ranks by title matches, then year.

Counts include every matching paper, not only the displayed page. Acceptance rates use the same phrase rules on all records in each verified ICLR submission pool. Both bars share that pool; published organizer rates are separate references. Rates require the Research track alone. Missing sources are not treated as zero.

No account, API key, analytics, or search-history storage is used. Matching runs in your browser. Index file requests can reveal the words being searched to the hosting provider; the complete query is not sent to a search service. Public data may be cached by your browser. GitHub receives file requests and maintains hosting security logs. Exporting all matching papers fetches their full metadata and can be a large download. A current browser with JavaScript, Web Workers and gzip decompression support is required. Chart exports are SVG/PNG; PDF and AI/semantic search are available in the full local app.

Metadata and abstracts retain their original rights and are not covered by the software license. Source links are provided with papers and rates. See data provenance and reuse notes. No paper PDFs or review text are bundled.