Policy: what we hold, and what we serve
zrxiv is a read-only mirror of the research corpus behind the Epistack cairns, operated by the Epistack project. This page says what we keep, what we show, and on what basis. Where a sentence depends on how this site is configured, it is rendered from that configuration.
What we hold
For each work we hold a record: its title, authors, year, DOI where it has one, and where it came from. For many works we also hold full text: the document we fetched and the text we extracted from it. We fetch full text only where upstream indexes report the work as open access. That is a test of access, not of licence: a work can be free to read and still carry no licence to redistribute it.
What we serve
The licence gate is on. Full text is served only where our tier map, the reviewed record of the licence on each copy we hold, confirms an open licence. Every other work we hold, including any work the tier map does not list, keeps its metadata page, its references and its chases, and its full text is withheld: its page says so and, where we have a DOI or an original link, points to where the work can be read, and the addresses of its files answer HTTP 451 (Unavailable For Legal Reasons).
The tier map, counted over the works on this site right now:
open— an open licence (CC BY, CC BY-SA, CC0 or public domain) on the copy we hold: 2,309 works. Full text served.open-nc— a non-commercial Creative Commons licence (CC BY-NC, BY-NC-SA or BY-NC-ND) on the copy we hold: 669 works. Full text served, with its licence and authors shown.open-nd— a no-derivatives Creative Commons licence (CC BY-ND) on the copy we hold: 10 works. Full text served, with its licence and authors shown.held— no confirmed open licence on the copy we hold: 3,809 works. Full text held, not served.clawback— a licence that lapsed: NLM withdrew the work from the PMC COVID-19 collection: 34 works. Full text held, not served.- not in the tier map: 7,128 works. Treated as held: full text not served.
These counts include works for which we hold no full text at all. Of the works whose full text we hold, 2,988 works are served here and 3,843 works are withheld.
Documents we write ourselves (chase documents, and the reference lists we extract from a paper) are not copies of anyone's work, and are served whatever the gate says.
Non-commercial and no-derivatives licences
Works under a non-commercial (NC) or no-derivatives (ND) Creative Commons licence are a separate class. With the gate on, they are shown on this free site with their licence and authors named, as those licences allow. If we ever offer a paid service, these works are excluded from what it reads until we have reviewed them.
Access is not a licence
That this site serves a copy of a work does not license you to reuse it. The work's own licence governs what you may do with it, and we cannot grant rights we do not hold. With the gate on, a served work's page shows its licence, and the JSON API and MCP endpoint report every work's tier and licence.
Held for research, not served
Text we hold without a confirmed open licence is kept, not deleted. We keep it for the Epistack project's own research, and it is not served from this site, its JSON API or its MCP endpoint. Withholding is a decision about serving, and it is reversible: a reviewed update to the tier map can serve a work whose licence we later confirm.
We do not get around access controls
We fetch only what a publisher or repository serves to an ordinary visitor. None of our fetching code solves challenges: there is no headless browser, CAPTCHA solver or bypass logic in it. Our fetcher sends a browser-like User-Agent because several open-access repositories refuse the default User-Agent of the programming library it uses, while serving the same file to a browser; the header changes how the client names itself, nothing else. When a Cloudflare challenge stood between us and a paper we wanted, we did not pass it: our own User-Agent and a browser's got the same refusal, and there it stayed.
No training
We do not train AI or machine-learning models on anything we hold.
If we got something wrong
If you believe this site serves something it should not, see takedown. For anything else, write to contact@zrxiv.dev. The terms of use are at terms, and what we record about visitors at privacy.