A preprint appears on a research repository in March, claiming something genuinely interesting. A journalist cites it. A policy analyst references it in a report. Four months later, the same underlying research goes through peer review, and the conclusions shift meaningfully — a caveat added, a claim narrowed, sometimes a finding reversed entirely. The people who cited the March version, in most cases, never find out the July version exists. Research doesn't stand still after it's first published. Most of the tools built to track it, though, treat a paper as a single, fixed, unchanging fact the moment it first appears.
Research Evidence Radar was built directly against that mismatch. Scientific and academic discovery has always been scattered across a genuinely large number of separate systems — arXiv, bioRxiv, medRxiv, Crossref, OpenAlex, Semantic Scholar, PubMed, and countless institutional repositories, each serving a different corner of the research landscape. Individually, several of these have become reasonably well-served by existing tools. What nobody had built was a way to follow a specific piece of research as it moved through its actual lifecycle — from an early preprint, through revision, through peer review, sometimes all the way to a formal published version, and, in the rarer but genuinely important cases, through to a retraction.
That lifecycle-tracking idea came from recognizing what the existing tools in this space were actually optimized for. A search of the marketplace found narrow, single-purpose extraction actors — respectable engineering, minimal adoption, and a shared blind spot: every one of them treated a paper as a static record to be pulled once, not an evolving piece of evidence to be followed over time. The demand for the underlying data was clearly real, evidenced by more established scraper categories in adjacent research domains. What was missing wasn't access. It was continuity.
Building continuity into a research-tracking tool meant solving a genuinely tricky identity problem before anything else could work. The same underlying research often exists in multiple places at once — a preprint on one platform, its eventual peer-reviewed version on a completely different journal site, sometimes under a slightly different title or author ordering. Without correctly recognizing that these are the same evidence at different stages, rather than two unrelated papers, any attempt at lifecycle tracking falls apart immediately. Getting that entity resolution right, reliably, across sources that don't share a common identifier system, became the single hardest and most important piece of the entire build — considerably harder than the extraction itself.
Once that foundation held, the more interesting layer became possible: tracking not just whether a paper exists, but how it's changed. A preprint published in March, peer-reviewed in June, and substantially revised in July isn't three separate events to note in isolation — it's a single evolving story, and telling that story clearly is what actually makes the tool valuable to someone trying to understand the current, most reliable state of the evidence on a topic. Retractions carry special weight in this system, flagged with real urgency rather than folded quietly into routine status updates, because very little damages research credibility as thoroughly as citing a retracted finding without realizing it's been withdrawn.
There's a particular audience for this tool that shaped its design as much as any human user did: research agents. As AI systems increasingly take on the work of synthesizing evidence and answering research questions, a clean, structured, continuously updated record of how evidence has evolved on a given topic becomes exactly the kind of infrastructure those systems need — not a static snapshot to reason from once, but a living feed built to be queried, trusted, and re-checked. Designing the tool's output to be genuinely useful to an autonomous research agent, not just readable by a human, became a deliberate part of the build rather than an afterthought.
The people who reach for this tool, human or otherwise, are trying to answer a version of the same question: what does the evidence actually say right now, and how confident should I be in it. A journalist fact-checking a claim wants to know if the study behind it has since been revised. A policy analyst building a report wants to know if new research has shifted a scientific consensus since their last review. A research agent synthesizing an answer wants a trustworthy, current account of where the evidence stands, not a static citation frozen at the moment it was first indexed.
Evidence, done honestly, is never really finished. It accumulates, gets revised, occasionally gets withdrawn entirely. This actor exists because tracking that motion, rather than pretending research holds still the moment it's first published, turned out to be the genuinely useful thing nobody had quite gotten around to building.