Primary Source
The original report of research or data: the study itself, the patent filing, the company's own disclosure. Reports built on primary sources are verifiable; reports built on summaries of summaries inherit everyone else's errors.
The vocabulary of literature reviews, technology landscapes, and research intelligence: what the terms mean and why they matter.
The first-pass review of a search's results, where a researcher reads titles and abstracts to decide which papers merit full-text review. In a systematic process, screening follows predefined inclusion and exclusion criteria so decisions are consistent and reproducible.
Alternative metrics of research attention, such as downloads, news mentions, policy citations, and social shares, used alongside traditional citation counts to gauge a paper's reach and influence.
A list of sources in which each entry carries a short note summarizing the work and assessing its relevance or quality. A useful intermediate step between a raw search and a finished review.
Finding relevant sources by working through the reference list of a known paper. A fast way to reach the foundational work a topic is built on.
A systematic error that skews a finding away from the truth, from how studies are selected to which results get published. Naming and controlling for it is central to trustworthy synthesis.
The AND, OR, and NOT logic used to combine search terms into a precise query, broadening or narrowing what a database returns.
Discovering relevant literature by following a paper's references backward (what it cites) and forward (what later cites it). Chaining catches important work that keyword searches miss.
Checking that a cited source actually supports the claim attached to it. This is the step most AI writing tools skip, and the reason many AI-generated reports can't be trusted. Atlas verifies every citation against the underlying source before a draft ships.
A structured analysis of the companies operating in a market: who they are, what they offer, how they're positioned, and where the white space is. Typically built from company disclosures, news, patents, and product literature.
Pulling predefined fields, such as sample size, methods, and outcomes, from each included study into a structured table. Doing it to a fixed template is what makes findings comparable across sources.
Removing duplicate records that appear when the same article is indexed in more than one database. A necessary cleanup step before screening so a paper isn't reviewed, or counted, twice.
A permanent, unique identifier assigned to a published article or dataset (e.g., 10.1000/xyz123). DOIs let a report link to the exact version of a source, which matters for verification and reproducibility.
A large biomedical and pharmacological literature database with strong coverage of drug research and European journals. Often searched alongside PubMed because each indexes sources the other misses.
Rating how much confidence a body of evidence warrants, based on factors like study design, consistency, and risk of bias. It tells a reader not just what the studies found, but how far to trust it.
The process of combining findings from multiple studies into a coherent overall picture, noting where studies agree, where they conflict, and how strong the combined evidence is. It's what separates a report from a reading list.
Finding newer papers that cite a given work, to trace how a finding has been applied, extended, or challenged since it was published.
The second-pass review of complete articles against the inclusion criteria, after abstract screening has narrowed the pool. The stage where borderline papers are finally kept or dropped.
A scatter plot of study results against their precision, used in a meta-analysis to spot signs that smaller or negative studies may be missing from the record.
A widely used framework (Grading of Recommendations, Assessment, Development and Evaluations) for rating the certainty of a body of evidence and the strength of the recommendations drawn from it.
Research and evidence published outside peer-reviewed journals: preprints, conference proceedings, theses, government and industry reports, white papers. Often the most current material in fast-moving fields, but it requires more careful vetting.
Tying an AI system's output to specific, verifiable sources rather than letting it generate unsupported text. Grounding is what makes a generated report checkable, and it is central to how Atlas works.
When an AI system produces plausible-sounding but false information, such as a citation to a paper that doesn't exist. The core reason ungrounded AI output can't be trusted in research without verification.
The degree to which studies in a synthesis differ in their populations, methods, or results. High heterogeneity is a signal that combining findings into a single estimate may be misleading.
A workflow design where a person reviews and approves an AI system's work at defined checkpoints rather than at the very end. In Atlas, a human approves the scope before research begins and the draft before it's delivered.
A measure of how often a journal's recent articles are cited on average. A rough proxy for a journal's prominence, but a poor guide to the quality of any individual paper within it.
The rules, set before screening begins, that determine which sources belong in a review (e.g., date range, study type, population, language). Defining them up front is what makes a review reproducible instead of arbitrary.
How a database catalogs an article, with metadata, keywords, and controlled vocabulary, so it surfaces in the right searches. Inconsistent indexing is a common reason relevant papers get missed.
A periodical that publishes peer-reviewed research articles within a field. The traditional unit of formal scientific publication.
A recurring meeting where researchers critically discuss a recent paper, its methods, and its conclusions. A working example of how findings should be scrutinized rather than taken at face value.
The ordering of journals by influence metrics such as impact factor or quartile. A rough, and often over-relied-on, proxy for the standing of the work inside them.
A term representing a concept in a search. The building block of a query, and the first place a search strategy succeeds or fails.
A question the existing evidence doesn't yet answer. Surfacing gaps is often the whole point of a review, pointing to where new research or investment should go.
A managed store of an organization's own documents and data. Atlas can combine a repository with public literature so reports reflect what the organization already knows alongside what the world knows.
A broad survey that maps the players, technologies, and activity across a domain to show its overall shape and where the white space lies, rather than answering a single narrow question.
An unanswered question or under-studied area that a review reveals. Identifying gaps is often the point of the exercise, pointing to where new research or investment should go.
A structured survey of published research on a defined question: what's known, what's contested, and where the gaps are. Distinct from a systematic review in rigor, but the best literature reviews still follow an explicit, repeatable search method.
Curated, validated information about a market, spanning competitors, customers, technologies, regulation, and trends, gathered to support a specific business decision rather than general awareness.
A statistical technique that combines quantitative results from multiple comparable studies into a single pooled estimate. The most rigorous form of evidence synthesis, and only possible when the underlying studies are similar enough to combine.
The documented approach behind a report, including how sources were searched, screened, and synthesized. A stated methodology is what lets a reader judge, and reproduce, the work.
A literature review that summarizes a topic through expert selection rather than a formal protocol. Readable and useful for orientation, but harder to reproduce than a systematic review.
Computational techniques for reading and analyzing human-language text at scale. The technology that lets research tools search, extract, and summarize across thousands of documents.
A finding of no significant effect. Scientifically valuable but chronically underreported, which is one way the published record ends up skewed.
Research published free to read without a paywall, widening who can verify a claim and build on the work.
A persistent digital identifier that reliably distinguishes one researcher from another across publications, even when names are shared or change.
The specific variable a study uses to quantify an effect. Evidence syntheses compare these across studies, which is why consistent definitions matter.
An analysis of patent filings in a technology area: who's filing, where activity is accelerating, and which approaches are being protected. A leading indicator of where companies are placing R&D bets.
The evaluation of a manuscript by independent experts before publication. Peer review is a quality filter, not a guarantee, which is why claims should still be checked against the paper itself.
A research paper shared publicly (e.g., on arXiv, bioRxiv, medRxiv) before peer review. Preprints surface findings months earlier than journals, at the cost of an extra burden of scrutiny on the reader.
The original report of research or data: the study itself, the patent filing, the company's own disclosure. Reports built on primary sources are verifiable; reports built on summaries of summaries inherit everyone else's errors.
A widely adopted reporting standard (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) that specifies how a systematic review should document its search, screening, and selection process, including a flow diagram of what was found and excluded.
The U.S. National Library of Medicine's free index of more than 37 million biomedical and life-sciences citations. The standard starting point for literature searches in medicine and biotech, and one of the primary databases Atlas searches.
Research producing non-numeric data such as interviews or open text, analyzed for themes and meaning rather than measured with statistics.
Research producing numeric data analyzed with statistics to measure size, frequency, or association. The kind of evidence a meta-analysis pools.
A structured search request submitted to a database, combining keywords, Boolean operators, and filters into a single expression.
The property that another researcher, following your documented method, would arrive at substantially the same result. In research reporting, it means the search strategy, criteria, and sources are explicit enough to be re-run and checked.
A managed collection of an organization's own documents and data, such as prior reports, internal studies, and proprietary datasets. Atlas lets teams combine their repository with public literature so reports reflect both what the world knows and what the organization already knows.
The explicit definition of what a research project will and won't cover: the question, the boundaries, the sources, the timeframe. Getting scope right before research starts is the cheapest quality control that exists, which is why Atlas asks a human to approve it first.
The explicit combination of keywords, Boolean logic, databases, and filters used to find sources for a review. Writing it down is what makes a search transparent and repeatable rather than ad hoc.
A work that describes or interprets primary sources: review articles, news coverage, analyst summaries. Useful for orientation, risky as the basis for a claim.
A literature review conducted to a formal protocol: predefined question, documented search strategy, explicit inclusion criteria, and structured screening. Designed to minimize bias and be fully reproducible.
A combined view of a technology area drawing on both scientific literature and market activity (publications, patents, companies, and products) to show where a field is and where it's heading.
The systematic search for emerging technologies, research capabilities, or potential partners outside an organization, typically to fill an innovation gap identified inside it.
A 1 to 9 scale for describing a technology's maturity, from basic principles observed (TRL 1) to proven in operation (TRL 9). Originally developed by NASA, now widely used in R&D and innovation scouting to compare technologies at a glance.
A review that synthesizes findings across multiple existing systematic reviews, sitting one level above them to summarize a whole body of evidence.
Information without a predefined format, such as free text, PDFs, or images, which must be parsed and organized before it can be analyzed.
Re-running a search to fold in newly published work, keeping a recurring report current without redoing it from scratch.
Confirming that a finding or dataset is accurate and fit for the use it's being put to, rather than accepted at face value.
A factor that is measured or manipulated in a study. Variables are the raw material of quantitative analysis and the terms in which results are reported.
A structured map of the suppliers or providers in a market, covering who they are, what they offer, and how they're positioned relative to each other.
An authoritative report that presents a position, technology, or set of findings, often published by a company to inform, and persuade, a technical audience.
An unmet need or unaddressed gap in a market or technology area. Finding it is frequently the goal of a competitive or technology landscape.
A preliminary research paper circulated for feedback before formal peer review and publication. Early access to ideas, at the cost of extra scrutiny.
Extensible Markup Language, the format used to encode article metadata and full text so databases and tools can exchange them reliably (for example, PubMed and JATS records).
Checking a claim against more than one independent source before relying on it, the habit behind trustworthy citation and a core step in how Atlas verifies reports.
Running a single query across several databases at once so that coverage doesn't depend on any one index, then deduplicating the combined results.
The publication date bounds set on a search, one of the criteria that keeps a review focused and makes it reproducible.
The number of records a search returns before screening. An early signal of whether a query is too broad, too narrow, or about right.
A comparison of activity between equivalent periods, used to spot trends in publications, patent filings, or funding over time.
A standardized statistic expressing how many standard deviations a value sits from the mean. Used when combining or comparing results across studies.
An open-access repository, operated by CERN, where researchers deposit datasets, software, and papers and receive a citable DOI for each.
A search that returns nothing. Usually a prompt to revise terms, spelling, or scope rather than evidence that nothing exists.
Atlas turns research questions into structured, citation-verified reports: literature reviews, technology landscapes, and company intelligence.
Start Research