Source support¶
“Supported” is separated into discovery, acquisition, and live maturity.
Run arp sources or arp sources --spec <corpus.yaml> for offline capabilities and per-venue/year gaps. See source inventory and acceptance criteria. No priority venue list has been selected; existing ACL/IJCAI/IEEE/ACM/ScienceDirect scenarios do not imply coverage of all conferences and journals. Enumeration completion is relative to the chosen source; upstream omissions remain possible.
Venue presets¶
Built-in venue presets let a spec name a venue by its short name, such as name: INFOCOM or name: IEEE TMC: the venue receives an identifier checked live in November 2026 (an ISSN for 47 IEEE journals, including the robotics journals T-RO, RA-L, T-Mech, RAM, and T-FR, for Wiley's Journal of Field Robotics, whose papers are manual downloads, and for AAAI, an OpenAlex series for INFOCOM, ICC, GLOBECOM, WCNC, VTC, ICDCS, ICNP, IWQoS, SECON, and SIGCOMM, and per-edition proceedings DOIs for MobiCom and MobiHoc). A weekly GitHub workflow rechecks every preset and official index. Identifiers written in the spec win, preset: none turns matching off, and arp sources lists every preset.
When OpenAlex lists far fewer works for an edition than for the year before, discovery keeps the slice complete but adds a possible_upstream_lag warning; when abstracts stay pending because a Semantic Scholar or OpenAlex key is missing, it prints the arp config setup command. Rerunning the same command into the same output reads warned slices again and looks up missing abstracts again; finished slices are not fetched again.
Sources¶
| Source | Discovery | PDF/BibTeX acquisition | Current maturity and fallback |
|---|---|---|---|
| Crossref | Primary venue/year enumeration and research search | No | Stable public metadata source; metadata completeness still varies by publisher |
| ACL Anthology | Official proceedings enumeration | Automatic | Mature and used in real corpus runs |
| IJCAI proceedings | Official proceedings and track parsing | Automatic | Mature and used in real corpus runs |
| AAAI (ojs.aaai.org) | Crossref venue enumeration with the AAAI ISSN; records carry abstracts and archive URLs | Automatic | Official OJS PDF and BibTeX export. Live-tested: AAAI-24 enumerated 2,867 records with abstracts; papers delivered. Since early November 2026 AAAI's citation export answers HTTP 520 for every client, so fetches end network_transient until AAAI fixes it |
| NeurIPS proceedings | Official yearly listing: main track and, from 2022, Datasets and Benchmarks | Automatic from discovery or the abstract URL | Official PDF and per-paper BibTeX file. NeurIPS DOIs resolve to Curran Associates, so a DOI alone is not enough. Live-tested: 2019 listed 1,428 papers with abstracts; papers delivered from 2019, 2022 D&B, and 2023 |
| ICLR proceedings | Official yearly listing on proceedings.iclr.cc from 2024 | Automatic from discovery or the abstract URL | Official PDF and per-paper BibTeX file. Earlier years are only on OpenReview, which answers non-browser clients with a challenge: discovery reports them source_unavailable and their papers are manual downloads. Live-tested: ICLR 2024 listed 2,260 papers with abstracts; a paper delivered |
| PMLR | Official volume BibTeX files of ICML, AISTATS, COLT, UAI, CoRL, ALT, ACML, L4DC, and CLeaR | Automatic from discovery or the paper URL | The BibTeX is the entry printed on the official page (publisher_page_embedded provenance). Live-tested: ICML 2023 listed 1,828 papers with abstracts; papers delivered from ICML 2019 and 2023 |
| CVF Open Access | Official CVPR, ICCV, and WACV main-conference listings | Automatic from discovery or the paper URL | CVF's open-access version, identical to the accepted paper except for the watermark; the final version is on IEEE Xplore (declare publisher: IEEE to keep that route). BibTeX printed on the page. Live-tested: WACV 2024 listed 846 papers with abstracts; papers delivered from CVPR 2019 and 2023, a CVPR workshop, and WACV 2024 |
| USENIX | Technical-sessions page of USENIX Security, OSDI, NSDI, USENIX ATC, and FAST, with each paper's abstract | Automatic from discovery or the presentation URL | Official open-access PDF and the BibTeX printed on the presentation page (publisher_page_embedded; the separate BibTeX download sits behind a browser bot check). Requests to www.usenix.org are spaced 10 s apart, as its robots.txt asks. Live-tested: OSDI 2024 listed 53 and USENIX Security 2023 422 papers, all with abstracts; OSDI and NSDI papers delivered |
| Springer Link | Crossref/OpenAlex/Semantic Scholar discovery; 10.1007 DOIs |
Automatic for journal articles and book chapters (LNCS) when the page states that this network has access | Official PDF and the BibTeX file from Springer's citation export. Open-access papers download everywhere, subscription papers only on an entitled network; otherwise access_required. No login. Live-tested: two Machine Learning articles and one ECCV chapter delivered |
| IEEE Xplore | Journals through Crossref (ISSN); conferences through OpenAlex (openalex_source, e.g. INFOCOM, ICC, GLOBECOM); abstracts from OpenAlex/Semantic Scholar |
Automatic with a configured CARSI institution profile | Generic profile-driven flow; live-tested end to end with one CARSI institution. Every additional institution requires first-use adaptation and a representative smoke test |
| ACM Digital Library | Usually Crossref collection enumeration plus optional query APIs | Conditional HTTP adapter | Works only when the official page and artifact endpoints allow the dedicated client. In November 2026 every non-browser request, and an automated Chrome, got a Cloudflare challenge, so ACM papers are manual items whose message names the bot check |
| ScienceDirect | Crossref/OpenAlex/Semantic Scholar discovery; Elsevier API can resolve expected identity | Manual only | The user downloads in a normal authorized browser; the CLI verifies and imports the local pair. No scripted ScienceDirect login or browser automation |
| Semantic Scholar | Optional keyed candidate search, DOI enrichment, and citation/related graph expansion | No | Discovery and graph evidence only; provider failures remain diagnostics |
| OpenAlex | Optional candidate search, DOI enrichment, and citation graph expansion | No | Requires its optional API key; discovery and graph evidence only |
| MinerU | No | No | Optional seven-day full-text parsing cache; it is analysis support, not a paper source |
| Generic direct publisher adapter | Internal host-bounded adapter exists | Not enabled as a public production default | Requires an explicit approved-host configuration and productized setup before it should be advertised as general support |