Screening
Screening, two blind votes at a time.
Two reviewers vote on every record without seeing each other. One of them can be AI. Conflicts get settled in the open, and the agreement gets measured.
The challenge
Thousands of records. Two tired humans.
The problem
A good search brings back a mountain. Screening it by hand is slow, and when reviewers can see each other’s votes, the second vote stops being independent.
With IAMB.AI
Blind by default, two votes per record, and an AI that can take one of them. You read the AI’s reasons only after you decide, so it helps without leaning on you.
What you get
Everything a screening stage needs.
Blind by default
Votes stay hidden from each other, and so do conflicts, until blinding is lifted.
Two votes per record
The default for every review. Change it if your protocol says otherwise.
An AI reviewer
An advisory vote, or a blind second reviewer at titles and abstracts and at full text.
Conflicts on the record
One list of disagreements. The owner’s resolution is kept with the record.
Cohen’s kappa
Agreement measured on each reviewer’s independent vote, taken before any discussion.
Full-text rules
An exclusion reason is required at full text, and a record can be marked not retrieved.
Auto-exclude, if you want it
Let the AI exclude records it is more than 95% sure about. PRISMA reports them as automation tools.
Reasons after you decide
The AI’s reasons, rationale and model show up once you have voted, never before.
A little XP
Optional game-style points for screeners. They never reward speed or a verdict. Your admin turns them on.
Screen the pile. Keep your weekends.
IAMB.AI is invite-only for now. Say who you are and what you want to research.