Config: the protocol of your review
Everything a review's Config tab holds, from the question and numbered criteria to blind mode, extraction mode and the autopilot's rules, and why each matters for the paper.
On this page
Config is the review's protocol in one place. Owners change it; the rest of the team can read it. Most settings feed the paper's methods section directly, so set them before screening starts.
The question
- Review question: one sentence.
- Research questions: they guide extraction (the AI proposes form fields that answer them) and the manuscript (the introduction states them and the synthesis answers them).
- PICOS: population, intervention, comparison, outcomes and study design, as the frame for your criteria.
Eligibility criteria
Write one criterion per line. Each line is numbered: I1, I2… for inclusion and E1, E2… for exclusion, and a line keeps its own label even if you reorder. Criteria are shown to screeners, sent to the AI with every screening prompt, and reported in the methods.
Why numbers matter: every screener, AI reviewers included, gives a verdict on each criterion (met, not met or unclear). A decision that contradicts its own verdicts is flagged in red, and the conflicts view marks the criteria on which two votes disagree, which is where a discussion should start.
Screening settings
| Setting | What it does |
|---|---|
| Reviewers per record | How many independent votes a record needs before it gets a final status. Two is the standard for dual screening (PRISMA 2020; Kitchenham's guidelines in software engineering). |
| Blind mode | Hides the other reviewers' votes, and the AI's suggestions depending on the reveal setting, so everyone decides independently. Conflicts can be seen and resolved only after blinding is turned off. The paper reports whether screening was blinded. |
| Exclusion reasons | Every full-text exclusion needs one reason; the PRISMA diagram lists the counts per reason. |
| Highlight keywords | Green for include terms, red for exclude terms in titles and abstracts. Use * as a wildcard: evaluat* matches evaluate, evaluated and evaluation. |
Extraction and related work
- Extraction mode: manual, AI pre-fill (people verify against quoted evidence) or AI automatic. See Data extraction.
- Related-work table: the dimensions on which prior surveys are compared. See Related work.
The team
Add people as owner, co-author, reviewer or viewer, and add AI reviewers with their own persona and model. See Roles and AI reviewers.
Protocol registration
If you registered the protocol (for example on PROSPERO or OSF), enter the registry and number. They are reported in the methods (PRISMA 2020 item 24).
Autopilot rules
When two AI reviewers run the review, Config also holds the rules for conflicts that discussion does not settle, and how many attempts a stage gets. The methods report them as rules, not as human decisions. See The autopilot.
Starting a review like a previous one? Clone the earlier review with setup only: Config, criteria and team come along, without records. See Clone, lock, archive and delete.