Glossary
The words of systematic reviews and of the app, in plain language, from AI co-author to worker computer.
A to C
AI co-author (AI reviewer). A member of a review's team backed by the AI, with its own persona and model. It screens, discusses, extracts and drafts, but cannot log in and is never an author. See AI reviewers.
AI-use statement. The paper's disclosure of what AI did and what people verified, generated from the job log. See The AI-use statement.
Archive. Keeping a review read-only for good once its paper is published, with its publication details. See Clone, lock, archive and delete.
Autopilot. A run that takes a review from the imported records to the PDF with two AI reviewers and no human step. See The autopilot.
Backward and forward citation searching (snowballing). Following an included study's references (backward) and the works that cite it (forward) to find more studies. See Searching the databases.
Blind mode. Screening without seeing the other reviewers' votes, so every judgement is independent.
Citation key. The short name of a reference in LaTeX, such as smith2025, used to cite it in the paper.
Cohen's kappa (κ). How much two screeners agree beyond chance: 1 is perfect, 0 is chance. See Agreement and Cohen's kappa.
Conflict. A record whose independent votes differ; it is discussed and resolved.
Co-author. A team member who screens, resolves conflicts, extracts and writes, and is listed as an author.
Criteria (eligibility criteria). The numbered rules that decide whether a study is in (I1, I2…) or out (E1, E2…).
D to K
Data extraction. Collecting the same pieces of information from every included study, with a form.
Deduplication. Finding and merging the same study found in several searches.
Economic nexus. In the startup guide: the level of sales into a U.S. state above which a seller must collect that state's sales tax.
Exclusion reason. Why a report was excluded at full text; the PRISMA diagram counts them.
Fidelity (records per prompt). How many records the AI judges in one prompt; fewer is more careful. See AI fidelity.
Full text. The whole report of a study, read to make the final eligibility decision.
Justification table. The comparison of prior surveys with your review on its key dimensions.
L to P
Librarian. The AI in the Library that answers questions only from a library's articles, with sources.
Main category. The category in which a study gets its full account in the paper (and its row in Table B).
Maybe. A screening vote for "not sure"; records everyone marked maybe go on to full text by default.
Merchant of record. In the startup guide: a payment service that sells your product in its own name and handles sales tax and VAT.
Owner. A team member with every right in a review, including Config and the team.
PICOS. Population, Intervention, Comparison, Outcomes and Study design: a frame for writing criteria.
PRISMA 2020. The reporting guideline for systematic reviews: a 27-item checklist and a flow diagram. See PRISMA.
PRISMA-S. The extension for reporting literature searches; your search log becomes its appendix.
Process record. Everything the paper reports, compiled from what the app recorded.
Protocol. The plan of a review, set before screening: question, criteria, methods. It can be registered (for example on PROSPERO or OSF).
Q to Z
Quality checklist. Questions about each study's rigor, answered yes, partly or no, giving a quality score.
Quote check. Looking up the AI's verbatim quote for each extracted value in the study's text.
Record. One search result: a title, authors, an abstract and other details.
Registered agent. In the startup guide: the person or company that receives legal papers for an LLC in its state.
Related work. Prior surveys of the same field, compared with your review.
Reviewer. A team member who screens and extracts.
Road. The eleven stages of a review shown across the top of the review on the public site.
Search log. Every search, run by the app or imported, with its source, date, query and counts.
Sensitivity and specificity. How many relevant records the AI kept (sensitivity) and irrelevant ones it excluded (specificity).
Systematic review. A review that finds, selects and synthesises studies by an explicit, reproducible method.
Table A and Table B. Per-category tables in the paper: approach and mechanism (A), results and cost (B).
Viewer. A team member who can read a review but not change it.
Worker computer. A computer that runs the AI engines (and on shared hosting, the paper builds and the jobs) for the site.