Methodology

From a new paper to a checked article.

Every central claim is tied to the primary paper. The process stops when access, evidence, or verification is insufficient.

Developing Light · Public record

01

Discovery

Crossref, Europe PMC, PubMed, and arXiv are queried through official APIs. The first online publication date matters, not a later index or deposit date.

02

Freshness and identity

DOI, PMID, PMCID, and arXiv identifiers are normalized. Papers from the first 72 hours receive priority; first discoveries older than seven days are not processed automatically.

03

Complete paper

Structured XML comes first, followed by repository HTML and openly reachable publisher PDFs. Provenance, access, license information, document version, and SHA-256 are recorded. An abstract alone is not enough for a full article.

04

Selection

Separate integrity, evidence, reader-value, freshness, and risk scores guide internal selection. They are working signals, not a public objective grade.

05

Analysis and evidence

Methods, sample, effects, uncertainty, limitations, funding, and status are captured in a structured form. Every central claim points to a section, page, paragraph, or table.

06

Independent verification

A fresh second pass reads the primary text and analysis again. Wrong numbers, missing evidence, an incorrect study type, or causal overstatement block the article.

07

Article check

The English draft may use only approved claims. A separate check reviews every text unit, especially the headline, introduction, and conclusion.

08

Autonomous release

A passing article is published automatically when its complete source text, source links, freshness, and independent verification satisfy the release gate. Every decision is recorded.

Publication status

Peer-reviewed means the paper is recorded as a reviewed journal publication; it does not make every claim automatically true. A preprint has not completed formal peer review. Corrections, expressions of concern, and retractions remain visible.

Earlier research

The primary paper remains the source for its own results. Reviews, meta-analyses, and guidelines provide context. No outside source may add a result value that the paper does not support.

Known limits

PDF extraction can lose tables and formulas. API metadata can be late or contradictory. Language models can make mistakes despite strict schemas. Low extraction quality, missing details, and ambiguous findings therefore trigger automatic reprocessing or exclusion.