← Research Radar

Privacy

Operated by [YOUR NAME OR ORGANISATION]. Contact: [CONTACT EMAIL]. Last updated [DATE].

What we store

• Your abstract or keywords, encrypted at rest, used only to scan for papers relevant to your project. Never shared, and never part of any dataset. Wherever Radar points at one of your sentences (the evidence under a flagged paper, the claims in "What's still new"), it stores only the sentence's position and reads the text from the encrypted abstract when you open the page.
• Your email address, if you sign in, to send sign-in links, the Monday brief and alerts you switch on.
• Your activity on flagged papers: the labels you give (threat, opportunity, noise), papers you save or open, and the scores Radar showed at the time. This tunes your own results and, only if you turn research sharing on, feeds the research datasets and the shared model described below.
• Your time zone and language, so emails arrive at a sensible hour.

What leaves our server

When a project is first scanned, a few short search phrases taken from it (for example "package registries" or "intrusion detection") are sent to OpenAlex and arXiv to find related papers. Your abstract itself is never sent to them or to anyone else. Emails are delivered through our email provider; the Monday brief names the papers that cover your claims but never quotes your abstract. We do not sell data, show ads or use your abstracts to train models.

Reading your full draft

If you give Radar your full draft (a PDF), your device extracts its text and sends it once. Radar picks out a few statements of the problem, method, data, contributions and results, keeps those encrypted together with the numbers it computes from them, and drops the text straight away; it is never written to disk or logs. You can read and correct the statements on the project page. The judge that explains each flagged paper runs on our own server: no outside AI service sees your draft.

Following people and labs

If you follow an author or an institution, Radar stores who you follow so it can show you their new papers. Nobody else sees your list.

Checking your references

When you check a manuscript's references, your device reads the PDF, .bib or text file and sends only the reference list. Radar compares it with the papers it holds in memory and keeps nothing: neither the list nor the result.

Learning from everyone

Each night Radar trains one shared model on the labels people give (threat, opportunity, noise), so that everyone's results improve, not only the labeller's. It follows the research-sharing switch below: only labels from signed-in people who turned research sharing on, and never from projects they leave out.

• What it learns from: for each labelled paper, the label and the nine or ten numbers Radar computed when it scored the paper (how close it is, how strongly it releases code, and similar). No text: no abstract, title, keyword or paper text goes into it.
• Safeguards: it is used only once at least 20 people and 300 labels are in, each person's labels count fully only up to 60 so no one person steers it, and it is switched on only when it predicts the labels of people it was not trained on better than Radar's fixed rules. Your own project's model, trained on your labels alone, still has the last word on your results.
• Turning it off: turn research sharing off and your labels are left out from the next night's training.

Checking a journal (Journal Trust Checker)

• Lookups. What you type (a link, ISSN, DOI or title) is checked against public registries and our record of verified cases. If you are signed in, your recent lookups are kept so you can see your history; visitors' lookups are not linked to them. A pasted email is read for its links and sender domains and then thrown away: its text is never stored.
• Daily limits. To count visitors' daily lookups we keep a salted hash of the network your address belongs to (not the address itself), and drop even that after 30 days.
• Fresh checks. When a site is unknown, our sandboxed fetcher visits it from our server, not from your browser. Captures of sites are kept as evidence; captures of sites that turn out not to matter are deleted after 90 days.
• Reports and disputes. If you report a site we keep that you reported it, so the people who verify cases can see who asked (and so you cannot verify your own report). If you dispute a result we keep your email, your statement and the outcome, to answer you and to keep the record straight; the dispute is deleted a year after it is decided. Verifiers' names are never public: the change log shows only their role.
• Publishers and institutions. Publisher claims keep the domain and how control was proven. Uploaded publication lists are parsed in an isolated process and discarded at once; we keep only the identifiers (DOIs, ORCID iDs, ISSNs, links, journal names) and the results, visible only to your institution. The database itself enforces that separation.
• The browser extension keeps its lists inside your browser and checks the sites you visit there. Only when the first four bytes of a site's hash match a list entry does it ask our server, and then it sends those four bytes, never the site or the page. It sends nothing about your browsing otherwise.
• Two-step sign-in and ORCID. If you turn on an authenticator app, its secret is stored encrypted. If you link ORCID, we keep your ORCID iD to let you sign in with it.
• Payments, when they are switched on, are handled by Paystack; we keep the reference, the product and whether it was paid, never card or mobile-money details.

Legal bases: providing the service you asked for (lookups, accounts, institution screening under the institution's agreement), our legitimate interest in protecting researchers from cloned journals (discovery, evidence, verification), and your consent where we ask for it. We act as data controller under Ghana's Data Protection Act, 2012 (Act 843) and the GDPR for people in the European Union. Ask for your data or its deletion at any time; we do it within 30 days.

Public research-area pages

The pages at /areas/ list new papers, released code and data, and open problems for research areas Radar watches. They contain public paper data only. An area gets a page only once several projects watch it, so a page never shows that one person is working on a niche topic.

Research datasets

Radar publishes research datasets once a month to support studies of how research fields move. Research sharing is off unless you turn it on. We ask every account once; you can turn it on or off at any time in Settings, and leave out single projects in their settings.

• What goes in: your labels, saves and opens on flagged papers, with Radar's scores, the papers' public IDs and titles, the research areas of your projects, and months (not exact dates).
• What never goes in: your abstracts, project titles, keywords, email address or anything else that names you.
• Safeguards: projects appear under pseudonyms that change with every release; labels are released only after 12 months; any group of fewer than 10 projects is left out or generalised; and the row-level label dataset goes only to approved researchers who accept a data use agreement forbidding any attempt to identify people or projects. Aggregate statistics are published openly.
• Why we may do this: your consent, given by turning sharing on. If EU or UK data protection law applies to you, this is our legal basis (GDPR Article 6(1)(a), with the research safeguards of Article 89). You can withdraw it at any time by turning sharing off, as easily as you gave it.
• Turning it off: you are left out of every release built after you turn it off. Releases already published stay as they are, because they cannot be linked back to you.
• Projects started without signing in are never used for research datasets.

Lab mode

If you join a lab and switch on sharing for a project, your supervisor sees the titles and labels of papers flagged for that project. They never see your abstract.

Cookies

We use one cookie to keep you signed in and, before you sign in, one to remember the projects you scanned. No tracking or advertising cookies.

Deleting your data

Delete a single project from its settings, or your whole account from Settings → Delete account. Deletion removes your projects, labels, activity record and email address straight away; database backups roll over within 14 days. You can also ask us for a copy of your data, or download your screening log from the Data page.

Paper data

Paper metadata comes from OpenAlex (CC0) and arXiv. Links point to the publishers' and arXiv's own pages.