Skip to content
Methodology

Source scoring methodology

Every source in the register earns 0–100 points. It is a practice score, not a truth score: it measures how open and consistent a source is — who stands behind it, whether it covers the big stories of its beat, how it treats its own texts. A high score does not mean a source is right; a low one does not mean it lies. The score means exactly what is written below, and nothing more.

The formula

The composite is the arithmetic mean of the criteria we can compute. Each criterion is 0–100 points earned from the components in the tables below. When a criterion cannot be computed for a source, it pulls the composite neither up nor down — the composite becomes the mean of fewer criteria, and we mark that.

Transparency

Who stands behind the source. Disclosure itself weighs most; the rest are verifiable traces: a named beneficial owner, a legal entity, a media-registry record, disclosed funding.

ComponentPoints
Ownership disclosure55
Beneficial owner named15
Legal entity on record10
In the media registry10
Funding disclosed10

Coverage completenesssoon

Viche's thesis measured on a single source: what share of its beat's big stories it covered, and which it stayed silent on. This criterion activates once the story-coverage aggregate exists — until then it counts toward no source's composite.

Editorial practice

Editorial behaviour: bylines, news–opinion separation, corrections practice, sourcing. Today this is a curated assessment from our research; where it does not exist yet, the criterion is marked unscored rather than zero.

ComponentPoints
Curated editorial assessment100

Signal reliability

The technical honesty of the stream: regular output, full texts, stable collection. Components that do not apply to a source (say, full-text share for a social-only source) redistribute their weight instead of zeroing the score.

ComponentPoints
Activity: articles past 7 days40
Full-text share30
Fetch health past 7 days30

What one source is

A source is a newsroom, not a feed. When an outlet publishes the same story to its site, its Telegram channel and its YouTube channel, that is one source with three channels: one score, one voice in an event's coverage. So «2 sources» under an event means two different newsrooms, not two feeds of one. Every channel is listed on the source's own page, and the register shows the count beside the code.

Bands

80–100exemplary practice
60–79good practice
40–59partly opaque
0–39opaque

Four honesty rules

  1. We do not score truth. The word “credibility” does not appear in this product: a source scoring 85 can be wrong, a source scoring 40 can be right.
  2. The formula is public. Every source's every point decomposes into components on its page. If the arithmetic cannot be shown, the number is not shown.
  3. Missing data is not bad practice. An unscored criterion is marked and excluded from the mean; it never counts as zero.
  4. No movement arrows until score history exists. “Better/worse than last month” arrives only with stored monthly snapshots.

Two languages

An event's summary, its key facts and a storyline's title are Viche's own words: a language model writes them after reading every report of the event. The English is written by the same model in the same pass, from the same reports — it is not a machine translation of the Ukrainian, it is the same material composed in English. Outlets' own headlines are never translated: the source list shows what each newsroom published, in its own language, linked to the original. An event covered by a single source has no Viche text at all, so on an English page it stays in Ukrainian and is marked as Ukrainian.

Where the data comes from

Components are computed from Viche's public read plane: researched ownership profiles (with their confidence level), the media registry, and 7-day collection metrics. Each source's page shows every input next to its points. Formula version: v1, August 2026.

Back to the source register