How Relativity Index Scoring Works
A calibrated, four-dimension relevance score, not a black-box include/exclude button.
The Relativity Index (RI) is a single, calibrated score that estimates how relevant a paper is to your review. It's designed to be a transparent decision-support tool, one you can inspect and override, not an opaque AI verdict.
The four dimensions
Every paper's RI score is built from four weighted sub-scores, each assessed by the AI directly against your review's topic and criteria:
| Dimension | Weight | What it measures |
|---|---|---|
| Topical Relevance (TR) | 74.2% | How closely the paper's subject matches your review's topic and keywords. |
| Methodological Relevance (MR) | 14% | Whether the paper's methodology fits what your review is looking for. |
| Contextual Relevance (CR) | 0% | Study context (population, industry, geography); currently tracked but not weighted into the score. |
| Keyword Alignment (KA) | 11.8% | Overlap between the paper's own terms and your review's keyword list. |
Topical Relevance dominates the score by design, a paper can be methodologically interesting but if it isn't about your topic, it shouldn't score highly.
Calibration
Each dimension's raw AI-assessed score is passed through its own affine calibration (d_cal = α + β·d_raw) before being weighted and combined. This corrects for the AI's tendency to cluster scores in a narrow band, spreading them out so the final Relativity Index lands on a calibrated 0–65 scale rather than the raw 0–100 you might expect. A score of 65 is exceptional, not the ceiling of a 0–100 percentage.
Tiers and thresholds
By default, Zynthia sorts papers into three tiers based on their calibrated RI score:
- RI below 26.7 → marked excluded
- RI 26.7–40.1 → marked manual review (uncertain)
- RI 40.1 and above → marked included
These are starting points, not final decisions. Every paper's tier is a suggestion; a human reviewer confirms or overrides it during screening.
The Inclusion Threshold slider
On the screening page, the Inclusion Threshold slider lets you move the include/exclude cutoff to fit your review's own standards, tighter if you're seeing too many marginal papers marked included, looser if you're excluding papers you think deserve a closer look. Moving it recomputes each paper's live tier immediately across your whole corpus, and is saved to your project so it's still there next time you come back. Papers you've already confirmed a decision on, by hand, keep that decision regardless of where the slider sits, the threshold only affects papers still awaiting review.
What you see per paper
Alongside the RI score and tier badge, Zynthia shows the four sub-scores, a short justification, and evidence quotes highlighted directly in the abstract, so you can see exactly what drove the score instead of taking it on faith.