How leader scores work

The short version

Every leader starts at 50 — a neutral score. As our system reads news coverage, it adjusts that number up or down based on how each leader is being portrayed. A score of 65 means recent coverage has been clearly favourable; a score of 35 means it has been clearly unfavourable.

This is a measure of media sentiment, not of honesty, competence, or morality. A leader can be scored highly during a period of flattering coverage and poorly during a scandal, regardless of the underlying facts. Treat the number as “how is this person being reported right now,” nothing more.

How a story becomes a score change

1. Extraction. When a new article is published, a language model reads it and identifies which politicians are mentioned.
2. Sentiment. For each leader, the model rates the tone of the coverage toward them on a scale from −1 (strongly negative) to +1 (strongly positive).
3. Weight. The model also estimates how central that leader is to the story, from 0 (passing mention) to 1 (the story is about them). A brief name-check moves the score far less than a front-page profile.
4. Storage. Each of these becomes a row you can inspect on the leader's profile page, under “Coverage that moved the score.”

Turning mentions into a current score

Once a day, each leader's score is recomputed from scratch as a time-decayed weighted average of all their tracked mentions:

• Every mention contributes its sentiment, multiplied by its weight.
• Older mentions count for less. We use a half-life of about 30 days — a mention from a month ago has roughly half the influence of one from today, and one from two months ago has a quarter.
• The weighted average of sentiment (a number between −1 and +1) is mapped onto the 0–100 scale, centred on 50.

The result is that scores move quickly when the news cycle turns, then drift back toward neutral if coverage goes quiet.

Trend arrows

The rising / falling / steady label compares the latest score with the score from roughly a month earlier, taken from the daily history snapshots. The trend chart on each profile is built from those same snapshots.

Where new leaders come from

If the extraction step finds a politician who isn't already tracked, they are added automatically at the neutral starting score of 50. Their score only becomes meaningful once a few weeks of coverage have accumulated.

Known limitations

• We track a defined set of news sources. A leader covered heavily by outlets we don't ingest will have a thinner, noisier score.
• Sentiment models can misread sarcasm, quotes, and headlines written for effect.
• Coverage volume differs enormously between countries, so scores are most useful relative to a leader's own history, not as a cross-country ranking of popularity.