Research does not end at publication.

For most of the modern history of science, the path from a published paper to a reader ran through a small number of legible channels: an index, a table of contents, a colleague, a reference list. Those channels were imperfect and often unfair, but they were at least visible. You could ask where a paper appeared and get an answer.

That is no longer how most encounters with the literature begin. A clinician, a reviewer, a policy analyst, or a graduate student increasingly starts by asking a question of a system, and receives an answer assembled from a handful of sources the system chose. Somewhere in that process, a great deal of good research is simply never reached — not judged and set aside, but never surfaced at all. And some of what is surfaced arrives with its population widened, its effect size firmed up, and its limitations quietly dropped.

Both of those are measurable. You can ask, systematically, whether a paper is retrieved for the questions it actually answers. You can ask whether it is cited once retrieved, and whether the citation supports the claim it is attached to. You can compare a summary against the study’s own stated population and endpoint and see where it drifted. None of this requires speculation about how any particular system works internally. It requires an instrument, applied consistently, and reported with its error.

The second thing worth saying is that most of what limits discoverability is dull and fixable. Terminology that matches the subfield rather than the question. An abstract that buries the population in a narrative paragraph. An unregistered identifier. A preprint never linked to the version of record. A paper with no lawful open version. These are not scientific failings, they are metadata and phrasing, and they are correctable — most cheaply before submission, when the text is still editable.

The uncomfortable part is that a company measuring visibility has an obvious incentive to make visibility look more measurable, more causal, and more important than the evidence supports. That incentive is why the product refuses to publish a public quality score, why simulated output is labelled everywhere it appears, why time-series changes are never described as effects, and why the research page currently lists intentions instead of results. If those constraints ever quietly disappear, that will tell you something.

This is a small, founder-led effort at an early stage. The measurement is a product heuristic under active validation, not a validated scientific metric. There are no published studies yet, no completed benchmark, no security certification, and no customer case studies — and every one of those absences is stated on the page where someone would look for it, rather than papered over.

What exists is a working instrument, a documented methodology, and a set of constraints written down early enough to be held to. Enterprise pilots are open. Individual plans are in private beta.