How Relaylit works
Methodology and limitations · Reviewed by the Relaylit team ·
What is Relaylit?
Relaylit is a literature monitoring tool. It searches six research databases, selects and ranks papers against your topic, and emails a digest. It helps you decide what to read; it does not replace a systematic review or assessment of a study's methods.
Sources and coverage
The sources are PubMed, Europe PMC, arXiv, Semantic Scholar, Crossref and OpenAlex. Their APIs may return incomplete records, index with a delay, or be temporarily unavailable. Searches have result limits: querying six databases does not guarantee finding every relevant paper. PubMed includes selected preprints through the NIH Preprint Pilot; Europe PMC and arXiv provide additional coverage.
From your brief to a shortlist
Your description is interpreted into interests, terms and exclusions for source queries. Duplicate records are consolidated using identifiers such as DOI and normalized identity signals. Filters and metadata help remove unsuitable records. Initial shortlisting combines topic matching, age and publication signals; not every retrieved result is sent to the AI.
What does the AI assess?
In each selection pass, the AI receives up to 35 candidates: title, venue, publication types and the first 400 characters of the available abstract. It estimates relevance to your interests and role. It does not necessarily read full texts or independently assess scientific quality. If this stage fails, the system can use heuristic ranking.
What does the score mean?
When AI ranking is available, ordering combines topical relevance, age and source or publication-type signals, with exclusions and feedback where available. Recency mode puts more weight on publication date. The visible 0–100 score is recalibrated within the selection: it is not a probability, an accuracy percentage or a scientific quality measure. Scores from different digests are not directly comparable.
Dates, preprints and summaries
Without an explicit age limit, the system may widen its time window to find useful material; an explicit age cap is retained. Preprints do not imply peer review. Generated summaries and notes depend on the available metadata and can contain errors: check the original paper before relying on a conclusion.
When is another option enough?
A native PubMed or Google Scholar alert may be enough for a simple, low-volume query. Systematic reviews require documented searches, screening and appraisal; a prioritized digest does not guarantee completeness. Relaylit is useful when you want to combine sources and prioritize recurring reading.
An example using real papers
Illustrative example, not a generated digest or an evaluation result. Actual selection depends on dates, filters and available search results.
Parameter-efficient adaptation of large language models: prioritize low-rank methods. Exclude papers focused only on attention architecture. Include preprints and older background work.
| Original paper | Fit to the brief |
|---|---|
| LoRA: Low-Rank Adaptation of Large Language Models | Matches the requested method; a candidate for reading. |
| Attention Is All You Need | Provides architecture background but does not address the requested adaptation method. |
No invented scores are assigned. Titles link to the original records so you can inspect the example.
Plans and delivery
Free: 2 active and 5 saved topics, weekly delivery. Pro: unlimited topics for USD 3/month or USD 30/year; weekly, biweekly or monthly delivery. Pro's daily manual generation allowance is separate from scheduled delivery. See all plan limits.
Sources and editorial responsibility
This page describes the implementation reviewed on the date above, not comparative trial results. The Relaylit team maintains this methodology; no external institutional validation is claimed.