Cross-checked, not single-sourced
Citations are the most-gamed number in science, so Frontier never trusts one index. It reads each work's citation count from up to five independent indices (OpenAlex, Crossref, Semantic Scholar, OpenCitations, Europe PMC): how many return one feeds the Evidence pillar, and their median and agreement feed confidence. Impact is anchored on field-relative measures (FWCI, the field-and-year percentile, and NIH iCite RCR) plus the pace of recent citations and Semantic Scholar's influential citations, never on a work's total citation count.
Beyond citations
Citations only measure what other papers did. A genuine breakthrough also reaches the world, so each result is corroborated against entry into Wikipedia, technical-community discussion, registered clinical trials, released datasets, and public funding. Each adds a few points to the Evidence pillar and raises a work's confidence: each is an independent line of evidence that the result is real and landing.
How freshness works
- Ingest. Every day, Frontier pulls new high-signal work from leading journals and preprint servers and adds up to 6: first papers their own journal ran a commentary on, then the highest-scoring of the rest, rotated across fields.
- Enrich. Each entry's sources are re-queried about once every 7 days, stalest first, so signals stay current.
- Score and trace. The score is recomputed, and each number decomposes to its raw source on the breakthrough page.
The source register
Scroll sideways for every attribute. Coverage is counted live.
| Source | Role | What we ingest | Data type | Timeliness | Update cadence | Coverage | Scale | License | Access |
|---|---|---|---|---|---|---|---|---|---|
| Citation & bibliometric | |||||||||
| OpenAlex | Bibliometric backbone | Field-Weighted Citation Impact, the field-and-year citation percentile, total + recent citations, the yearly citation trajectory, work type, open-access status, retraction flag, reference count, authors, institutions | Scholarly metadata + metrics | Days to weeks after publication | Continuous; new work daily, every score re-checked weekly | 390 of 390 | 250M+ works | CC0 1.0 | REST API, keyless |
| Crossref | Registration + independent citations | is-referenced-by count, reference count, funder count, reuse license, and retraction/correction notices (update-to and updated-by) | DOI registration metadata | Real-time on deposit | Continuous; weekly refresh | 387 of 390 | 160M+ records | Open metadata | REST API, keyless |
| Semantic Scholar | AI-enhanced citations | Citation count and influential-citation count (citations that genuinely build on the work rather than mention it in passing) | AI-curated corpus | Weeks | Continuous; weekly refresh | 370 of 390 | 200M+ papers | ODC-BY | REST API, keyless |
| OpenCitations | Open citation index | An independent, fully-open citation count: a fourth index cross-checking the citation figure | Open citation data | Weeks to months | Periodic releases; weekly refresh | 390 of 390 | 2B+ citation links | CC0 | REST API, keyless |
| Europe PMC | Life-sciences literature | A fifth independent citation count, PubMed/PMC identifiers, and open-full-text + linked-dataset flags | Biomedical literature | Days | Daily; weekly refresh | 329 of 390 | 45M+ abstracts | Open / per-record | REST API, keyless |
| NIH iCite | Field-normalized biomedical impact | Relative Citation Ratio (RCR), Approximate Potential to Translate (APT), citations from clinical articles, and the field citation rate | Bibliometric analytics | Rolling, annual RCR | Monthly; weekly refresh | 289 of 390 | 30M+ PubMed articles | CC0 | REST API, keyless |
| Open science | |||||||||
| Unpaywall | Open access + licensing | Open-access status and route, the reuse license (CC-BY, CC0), whether a green repository copy exists, and the host type | Open-access location data | Days | Continuous; weekly refresh | 347 of 390 | 30M+ OA works | CC0 | REST API, email-keyed |
| DataCite | Released datasets + software | How many open datasets and software artifacts (across Zenodo, figshare, Dryad, Dataverse, and institutional repositories) are formally linked to the work: reproducibility depth | Research-artifact registry | Days | Continuous; weekly refresh | 43 of 390 (works that released data) | 60M+ DOIs | CC0 | REST API, keyless |
| arXiv | Preprints (the leading edge) | For work whose record is an arXiv preprint (or whose arXiv id was added by hand): its version/revision count (a proxy for active iteration) and its primary category. A journal paper's own preprint is not looked up | Preprint server | Same day | Daily; weekly refresh | 9 of 390 (preprinted work) | 2.4M+ preprints | Per-preprint (many CC) | OAI/REST, keyless |
| Real-world reach | |||||||||
| Wikipedia | Entry into general knowledge | How many Wikipedia articles cite the work by DOI: whether a result has crossed from the literature into the general encyclopedia, which heavily-cited-but-obscure work never does | Encyclopedic references | Minutes (live wiki) | On refresh; weekly | 68 of 390 | 7M+ articles (English Wikipedia) | CC BY-SA | MediaWiki API, keyless |
| Hacker News | Technical-community discussion | Stories, points, and comment volume for the work (matched on the arXiv URL), a real-time signal of whether the practitioner community actually noticed and argued about it | Community discussion | Real-time | On refresh; weekly | 2 of 390 (arXiv-posted work) | 40M+ items | Public API | Algolia API, keyless |
| ClinicalTrials.gov | Clinical translation | Registered human trials that cite the work by PubMed id, and the highest trial phase reached: whether the result progressed toward the clinic. Correctly reads zero for non-clinical work | Trial registry | Daily | On refresh; weekly | 30 of 390 (clinical work) | 500K+ studies | Public domain (NLM) | REST API v2, keyless |
| NIH RePORTER | Public research funding | Which US federal (NIH) grants funded the work, via its PubMed id, and their total award dollars: the public investment that backed it | Grant / funding data | Monthly | On refresh; weekly | 69 of 390 (NIH-funded work) | 2.7M+ projects | Public domain | REST API, keyless |
Coverage is the number of the 390 tracked breakthroughs whose stored data currently carries a record (or, for Unpaywall, DataCite and the reach sources, a positive hit) from that source; OpenCitations answers every DOI, with a zero when it holds no citations. A source contributes to a breakthrough only where it has a record; the Frontier Score falls back gracefully when a signal is absent.