Methodology & data sources

PubScope is built on a single rule: every signal is a verifiable fact drawn from a third-party record — never our own opinion, and never a guess. This page explains exactly how the Trust Score is computed, where each fact comes from, and how to correct anything you believe is wrong.

The Trust Score (0–100)

The Trust Score is a credibility signal, not a prestige or impact ranking. A journal is placed by its strongest verifiable indexing / vetting credential; holding several credentials adds a little more. Two principles keep it fair:

• Absence is never a penalty. Not being in a selective index is neutral, not negative — the major indexes systematically under-cover open-access, non-English and regional journals, so absence tracks access model and scope, not quality. (This follows the methodology of the OECD/JRC composite-indicator handbook and national registers such as Finland's JUFO and Norway's Kanalregisteret.)
• Access model is never scored. A paywalled flagship and an open-access journal with the same credentials score the same. Being open access is neither rewarded nor penalised.

Only integrity flags subtract. The bands:

Every journal page shows the full breakdown — the exact credential that set the score, the corroboration, and any penalty — so you can see precisely how the number was reached.

Where the facts come from

We aggregate open and third-party scholarly data. We do not maintain our own blacklist, and we do not invent metrics. Coverage and freshness by source:

The Journal Impact Factor is Clarivate's licensed metric; we link to the official Journal Citation Reports rather than reprinting a number that can be out of date. Metrics can lag their source — always confirm a career-critical fact against the journal's official record.

Integrity signals — facts, not verdicts

We surface a DOAJ withdrawal(with DOAJ's own stated reason and date) and retractionrecords as neutral, sourced facts. A withdrawal is not a predatory verdict, and we do not label any journal "predatory" from our own judgement — where the word appears it is a quoted attribution to a third-party record. There is no PubScope blacklist.

The no-fabrication rule (AI features)

The Ask Advisor, the Journal Finder explanations, and the Paper Review report are built so they cannot state a fact that isn't in a tool or database result.When the evidence isn't there, they say "not found" rather than inventing an impact factor, an editor, or a citation. The free Reference Check verifies each reference against Crossref, OpenAlex, DataCite and Semantic Scholar, catches retractions and wrong/AI-hallucinated DOIs, and reports a reference it cannot match as "not found" — never as "verified".

How each AI feature works

Journal Finder. A multi-stage ranking pipeline: a scholarly embedding model finds semantically nearest journals, keyword and semantic signals are combined and fused, and a relevance model re-ranks candidates with a Trust-Score weight. The match percentage is an absolute topical-fit signal (how far the abstract sits above a per-query baseline calibrated against the full catalog) — not an acceptance probability. It is strongest for distinctive-domain clinical and STEM abstracts and weaker for humanities/interdisciplinary work; when nothing fits well it says so rather than dressing up a weak match.

Ask Advisor. A chat grounded in the 46K-journal database via read-only lookups; it answers only from what those lookups (and web search for public facts) return, audits its own claims, and declines to invent a journal, metric or editor. Any memory it keeps is a derived fact for your account only — your manuscript is never stored.

Paper Review. A pre-submission report, in two parts: (1) a deterministiclayer — every reference checked for existence, retraction, and DOI-vs-title mismatch against Crossref, OpenAlex, DataCite, Semantic Scholar and Retraction Watch; plus statistical consistency checks (sample size vs. significance, percentage/count mismatches), readability metrics, numerical consistency across figures and text, and a related-work coverage check against OpenAlex's most-cited recent papers; and (2) an LLM reviewer panel (methodology, novelty, clarity and a devil's-advocate lens) whose every finding must quote a line from your manuscript— ungrounded findings are dropped by an audit step. Results carry a readiness score (0–100) computed by a fixed formula from those checks — mostly verifiable counts (citation resolution, structure, statistics), plus a rough topical-overlap signal when you name a target journal — and a prioritised revision roadmap. It is not a pass/fail verdict and not an acceptance prediction. Your manuscript is processed in memory and never retained.

Free Reference Check.Deterministic, no LLM: paste a reference list (plain text, BibTeX, or RIS) and each reference is resolved against the databases above; retractions and wrong/AI-hallucinated DOIs are flagged, and an unmatched reference is reported as "not found", never as "verified" or "fabricated". A standalone claim-verification endpoint also checks whether a cited paper's abstract actually supports the claim it is cited for. We publish a labelled test set and the measured result for this checker, including the reference list so you can reproduce it.

Compare. Side-by-side comparison of up to four journals: trust score, quartile, indexing coverage, APC, review time and open-access status — all drawn from the same verified database.

Watch.Follow journals you care about; when their indexing status, quartile, or trust signals change in a monthly data refresh, the change appears in your Radar feed, linked to the sourced record that changed. (Email digests of these changes are paused while we work out a sending cadence that isn't noisy — the in-app feed is live.)

Found something wrong? Corrections & appeals

Because every signal is sourced, an error usually means an upstream record is stale or a match is wrong — and we want to fix it. Every journal page has a "Report an error" link that opens a pre-filled form, or write to [email protected]. We correct verifiable mistakes.

Honest limitations

• The Journal Finder is strongest for distinctive-domain clinical and STEM abstracts; recommendations for humanities and interdisciplinary work are less reliable, and a very short abstract gives it less to work with. Treat its list as a starting point, not a verdict.
• Index coverage is never complete — a "Limited verification" score means we found no signal, not that a journal is illegitimate.
• Metrics carry the freshness of their source; we show as-of context where we can and link to the authoritative record for career-critical checks.