Research judged on substance, not signal

Mission

PubVerse exists to change how new research is judged. For most of the history of science, a good idea has had to wait for the crowd to catch up before anyone treats it as important. We think the substance of a result can be recognized on the day it appears, and not years later. That belief is the whole point of this work.

PubVerse, alongside fortytwodegrees.org, points the waymark of emerging knowledge and enlightenment. We quantify hidden relationships within knowledge spaces so ideas are evaluated on substance, not signal, and the ones that matter are recognized sooner.

PubVerse is built alongside fortytwodegrees.org and shares its aim of pointing toward emerging knowledge before it becomes obvious. The compass you see turning while PubVerse is at work is the same waymark.

What PubVerse is

PubVerse reads a research abstract and tells you three things about it: how new the idea is, how sound the methods look, and how much the work is likely to matter. It reaches those judgments by weighing the abstract against the published scientific literature in its field, not against a popularity score.

The guiding idea is substance over signal. Signal is everything that surrounds a paper and has nothing to do with whether it is right or important: the name of the journal, the size of the lab, the number of people already crowded around the same fashionable question. Substance is the work itself. PubVerse is built to look past the signal, read the substance, and say what it sees in plain language.

You get numbers, but the numbers are not the product. Each result comes with a short written verdict that explains, in ordinary terms, why the work reads as new, or careful, or likely to matter. The reasoning is meant to be read and argued with, the way you would argue with a thoughtful colleague.

How every judgment is grounded

Every score PubVerse produces is traceable to specific prior work retrieved from the database. When PubVerse says a paper is novel, it cites the nearest prior publications it compared against. When it says the methods are sound, it points to the prior art it used as the benchmark. No judgment comes from a model's internal knowledge alone.

The process works in three steps. First, PubVerse retrieves the most relevant prior work from its corpus of more than 116 million publications using semantic search over the full text of the scientific record. Second, it extracts the substance of each retrieved passage so the comparison is about findings and methods, not keyword overlap. Third, a local language model reads the retrieved evidence alongside the candidate abstract and produces a grounded verdict with citations to the specific prior work behind each claim.

This means every result has receipts. On the Compass page, each needle carries an expandable section showing the prior work it was evaluated against, how close each prior sits in similarity, and a short passage describing what that prior did. An expert can verify or challenge any call in seconds by reading the same evidence the model used.

When PubVerse cannot find relevant prior work in the database, it says so honestly. A paper whose contribution is absent from the retrieved work is not automatically scored as highly novel. PubVerse withholds its judgment rather than inflating a score on an uncontested claim, because the absence of a retrieved pre-emptor does not prove novelty. It may simply mean the relevant prior is not in the corpus or was not retrieved.

How to score an abstract

Scoring is the quickest way to see PubVerse at work, and it lives on the landing page.

Paste the abstract into the box. A live counter shows your word count against the limit. If you run over, the counter turns red and the Score button waits until you trim it. Press Score. PubVerse first makes sure the text is a genuine research abstract and quietly turns away posters, fragments, and writing that is not research at all. If the text clears that check, PubVerse evaluates it and returns three measures, each on a scale of 0 to 10: Novelty, Methods, and Impact. Below them sits a short verdict that puts the three numbers into a sentence or two of plain reasoning.

You do not need a title, authors, or a DOI to score an abstract. Those are asked for only if you decide to promote a result. Promoting adds the identifying details, checks that a DOI you supply really belongs to the work, and gives you a link you can share with anyone. Everything you score is kept in your own private history so you can return to it, and you can delete any of it whenever you like.

How the Compass works for you

The Compass is for watching a whole area of research rather than a single abstract. It takes its name from the 42 compass: a way to keep your bearings while a field moves.

Name a topic you care about. PubVerse confirms that it can cover that area, and tells you plainly if it cannot. Then you choose how far back to look, from the last month out to the last two years. PubVerse sweeps the recent literature in that area and hands back a short list of needles, the results worth your attention, each carried with the same three measures and a plain verdict. It also shows how much material it read to find them, so you can see the size of the haystack behind the short list.

You can save a topic to your profile so it is one click away next time, leave an email address to be told when a sweep has finished, and export a simple list of what the Compass found, either the needles alone or the full scope it searched.

What is a needle

A needle is a result that is both genuinely new and genuinely sound, found while it is still easy to miss.

Most research, by its nature, extends what is already known. That work is valuable, but it is not a needle. A needle breaks new ground, and it holds up when you look at it closely. The hard part is that these two qualities rarely announce themselves early. A truly new result often has few citations, sits in an unfashionable corner, or comes from a group no one is watching yet. By the time the wider field agrees it mattered, the moment to have acted on it early is gone.

The Compass exists to catch needles in that early window, when recognizing them is worth the most. When PubVerse calls something a needle, the verdict says in plain words why: what is new about it, and why the work looks trustworthy rather than merely surprising.

Reading the map

Above the needles is a map of the field itself. It is not a picture of the papers PubVerse judged. It is a picture of what the field has been working on, built from every preprint posted in the window you chose, with your needles placed on it.

Each shape is a cluster: a group of papers that talk about the same thing. The clusters are found by the same code that groups papers inside the scoring pipeline, so a cluster here means what a cluster means everywhere else in PubVerse. Every cluster is named by reading its own papers. Clicking one lists the papers that sit closest to its centre, which is the quickest way to check whether a name is fair. Hexagons are methods, triangles are impact, diamonds are open problems, and the small dots are terms that two clusters share, which is how one part of a field connects to another. Your needles are the red stars, and papers that scored above the run average are amber triangles.

The field fit band

The band asks how far inside the established structure of the field a paper sits, and answers in standard deviations.

Every paper is compared to the cluster it most resembles, which gives a similarity score. That number alone means very little, because clusters differ: some are tight and specific, others broad and loose, so the same score means different things in different parts of the field. So the score is measured against a reference. When the landscape is built, PubVerse records what a random paper from that field scores against each cluster, and how much that varies. A paper's band value is how far above that reference it falls, counted in standard deviations of the reference itself.

A value near zero means a paper looks about as related to its nearest cluster as any paper picked at random would. Two standard deviations means it sits squarely inside recognised work. The slider hides above-average papers below whatever value you choose, so dragging it right leaves the papers most firmly inside the field's existing structure, and dragging it left brings back the ones sitting further out.

Three things the band is not. It is not a model output: no scoring head is involved, and nothing here comes from the novelty model that judges your abstracts. It is arithmetic on where a paper sits relative to its neighbours, and nothing more. It is not a quality score, and it is not novelty. A paper can fit the field closely and be dull, or sit at the edge because it is unusual or because it is simply off topic, and the band cannot tell those apart. We measured the relationship between the band and the judged novelty of a paper and found it weak but real, so it is a soft ordering rather than a test. Treat it as a way to change what the map emphasises, not as a verdict.

Needles are drawn at every setting of the slider. That is a deliberate choice and it is worth understanding, because it means the fact that your needles survive a high cut says nothing at all. They would survive any cut, since the slider never removes them. What the readout tells you instead is how many of the above-average papers are hidden at that setting, which is a real number.

Network depth

The depth slider controls how far out from your needles the map is drawn. At one degree you see only the needles and the clusters they belong to. Each step outward adds the clusters that those clusters resemble, until at the last position the whole field is on screen. Turning it up shows more of the field and makes the picture busier without adding information about your own papers, so it is best used to get a sense of where your results sit rather than to read individual names.

The map opens two degrees out, which is usually enough to show the needles with the structure around them. Drag to move, scroll to zoom while the cursor is over the map, and use Expand or Full screen when you want room to work. Cluster names appear as you zoom in, and the small (i) beside each needle in the list below flies the map to that paper.

Affiliation