Skip to main content
Methodology

How our free tools work

We document everything our free tools do (and don’t do) so you can decide whether to act on a result. No black boxes; no hand-wavy “AI says.”

Data handling

  • Your text is not used to train models. Standard Anthropic API retention may apply. OpenAI is not used by these tools today.
  • Result caches are time-limited. Cached results are keyed by content hash and expire after 7 days for journal fit, and after 24 hours for citation checks, stats audits, and reference-integrity scans. A cache expiration is not a deletion promise for a separately saved tool-run record.
  • The Citation Claim Checker does not save raw claim or reference text in tool history. Its operational record contains input lengths, resolved public-paper metadata, source coverage, the outcome label, and the verified source passage. Raw inputs remain only in the 24-hour response cache and Anthropic processing path. The checker no longer emails raw claims or creates public result links.
  • No account or email gate. The free tools work without sign-up. Rate limits are per IP, not per account.

Journal Fit Predictor

The tool checks that the input is a real abstract, loads a fixed, disclosed corpus of 1,321 venues, and asks a language model to propose five candidates by topic overlap with your title and abstract. The server rejects any candidate whose quoted evidence is not an exact phrase in that venue’s corpus record, so every result shows the passage behind it.

It does not calculate acceptance probability, predict editorial decisions, evaluate manuscript quality, or label venues safe, realistic, or aspirational. It has not been benchmarked against editor shortlists or submission outcomes.

Read the Journal Fit Predictor methodology, corpus and data dictionary →

Citation Claim Checker

Evidence reviewed · Verify v2.0

Paper search

A DOI is resolved directly through Crossref. When Crossref has the identifier but no abstract, the checker attempts to enrich the exact DOI record with Semantic Scholar metadata. For an author-year or numbered citation without a DOI, the checker searches Semantic Scholar, Crossref, and PubMed through NCBI E-utilities in parallel. Candidates must match the author and publication-year window, then are ordered by overlap between claim or reference terms and the candidate title or abstract. The interface labels these as candidates rather than exact matches.

Source-text gate

The checker requires at least 200 characters of abstract text or at least 500 characters of eligible open-access full text before asking the model for an assessment. For every DOI-backed record, it attempts full-text retrieval first from PubMed Central and then from an eligible HTML location identified by Unpaywall, even when an abstract is available. Extracted body text may be truncated at 24,000 characters. Tables, figures, reference lists, PDF-only copies, and supplementary files are excluded. If neither source-text gate passes, the result is Unable to Verify.

Verdict

When a paper passes the source-text gate, we send the available text plus the claim to Claude Sonnet 5.5 with explicit scope and outcome definitions. For Supported, Partially Supported, and Not Supported, the model must also return an exact passage of 20 to 500 characters and identify its source. The server normalizes typography and whitespace, then checks that the passage occurs verbatim in the retrieved abstract or full text. If it does not, the checker changes the result to Unable to Verify. The five outcome labels are:

  • Supported. The claim is supported by the retrieved source text and includes a server-verified passage.
  • Partially Supported. The paper addresses the topic but the claim overstates, oversimplifies, or misses a qualifier.
  • Not Supported. The paper addresses the same topic but contradicts or does not support the specific claim.
  • Wrong Paper. The search returned a different paper, usually one by an author with the same surname. The claim itself may be fine.
  • Unable to Verify. No paper, insufficient source text, no usable passage, or a returned passage that fails the server-side verbatim check.

Retraction signal

When the matched record includes a DOI, the checker looks up Crossref retraction metadata, including retraction relationships on the work record. Crossref also incorporates Retraction Watch records, as described in its official retraction-data documentation. The check is best effort: missing or incomplete DOI metadata can produce no flag, so an unflagged result is not proof that a paper is current.

What this tool does NOT do

  • It does not retrieve paywalled full text, evaluate tables, figures, or supplementary files, or search an entire manuscript.
  • Its Crossref retraction signal is not an exhaustive integrity review and may miss records with incomplete or absent DOI metadata.
  • Author-year matching can return the wrong paper, particularly for common surnames. Add a DOI or full reference when available and inspect the matched title before using the verdict.
  • The model-generated assessment is not presented as a calibrated accuracy estimate, a human review, a determination that a claim is universally true, or a substitute for reading the source.
  • It is rate-limited at 20 checks per hour per IP. There is no client-side three-check paywall.

How to cite these tools

DOI10.5281/zenodo.19831122GitHubManusights/compass-corpusLicense: CC-BY-NC 4.0

If you report results from these tools in a manuscript, methods section, or supplementary materials, cite the tool and version you used. Cite the Zenodo DOI only for the archived 1,030-venue corpus (v1.1.0).

Manusights. (2026). Journal Fit Predictor (version 2.0, corpus edition 2026.04.27-live2) [Web application].
  https://manusights.com/tools/journal-fit (Accessed: YYYY-MM-DD)

Manusights. (2026). Compass v1.1: Journal Fit Predictor (Corpus + Methodology).
  Zenodo. https://doi.org/10.5281/zenodo.19831122  [archived corpus only]

Manusights. (2026). Citation Claim Checker v2.0 [Free academic tool].
  https://manusights.com/tools/citation-claim-checker
  (Accessed: YYYY-MM-DD)

Reference Checker

Retractions + hijacked journals + expressions of concern

The Reference Checker’s methodology is documented separately because each verdict cites a distinct authoritative data source: Crossref for resolution, Retraction Watch (ISSN 2692-4579) for retractions and expressions of concern, and the Retraction Watch Hijacked Journal Checker for impostor venues. The full disclosure (resolution chain, refusal cases, known limitations, source attribution per flag) is on a dedicated page so librarians can link from LibGuides without sending users through marketing copy.

Read the Reference Checker methodology →About + data-source credits

Want a manuscript-level signal, not a paste-level one? The full readiness scan reads your entire manuscript: it checks citation risk against scholarly sources, scores desk-reject risk for your target journal, names reviewer-flag patterns, and produces a prioritized fix list. Free preview, $39 for the full report only if you want it.

Run the full readiness scan