Skip to main content
FreAlgo
Menu

Methodology

How sources are chosen, how outlets are labelled, how stories are tagged, and how to ask for a correction.

Source orientation labels

Labels describe an outlet's general editorial orientation as assessed by a cited methodology on a stated review date. They do not describe the accuracy of any individual article, and they are never used to infer whether a story is true. An article from an outlet labelled “Center-Left” is not itself classified; article-level labels are a separate field that stays empty unless a story has been reviewed on its own.

Orientation is a directional spectrum with five positions. Three further states are deliberately not points on that line:

How a position is drawn (text + marker, never colour):

Words like liberal and conservative mean different things in different countries, so they are stored as separate descriptive tags with the national context they apply to, never mapped onto the spectrum.

Label status and confidence

Every rated label must cite a source URL and methodology; reviewed labels must carry a review date. Confidence (0–100%) expresses how strongly the cited evidence supports the label. Any label produced by an automated classifier is marked “machine-generated” and can never be shown as reviewed. Every change is kept in an append-only label history.

Source selection

Sources come from a curated registry recording publisher, country, primary language, feed URL, usage terms and label provenance. Only official RSS/Atom feeds or licensed APIs are used. Real publisher feeds start paused until their terms have been reviewed. Coverage is uneven across countries and languages, and the interface says so wherever it applies. See the source registry.

Country and sector tags

Tags are assigned by deterministic, versioned rules, and each tag records why it was applied:

Sectors (a story may carry several):

Rule-based tags can be wrong. They are aids for filtering, not editorial judgements.

Time and language

The publisher's original timestamp is kept alongside a UTC-normalized time. When a feed gives no usable date, the time we first saw the story is used and marked as estimated. Languages are as declared by the publisher. Nothing is machine-translated.

Corrections, appeals and takedowns