How Protagoras works

What we measure and what we don't

• Which outlets covered a story, and which did not.
• How often an outlet writes about a person or a topic.
• How it presents the government: with praise, with criticism or neutrally.
• Which political positions it supports in its editorials and in its regular columnists' texts. This is what the map shows.

We do not measure whether an outlet is reliable, "good" or "bad", or whether it tells the truth. Ownership is shown separately and does not change an outlet's place on the map.

The map

Politics does not fit on one line from left to right. An outlet can ask for more state in the economy and at the same time hold conservative views on social questions. That is why each outlet is a dot on a map with two axes, on a scale from −3 to +3.

We measure five axes. The map opens on the economy and society. The three additional ones appear only once they have enough texts and pass their checks; then you can choose any two of the five.

The parties are placed as the outlets are, from their official texts (programmes, decisions, speeches), and help you "read" the map. An outlet being close to a party does not mean it "belongs" to it.

How a dot gets to the right place

1. We set the issues in advance. A published list of 46 statements for the five axes, such as "The state should intervene in the housing market, with rent limits or social housing". Which side support counts towards is also set in advance.
2. We read those who speak for the outlet: its editorials and its regular columnists, over the last 12 months. Guest opinions, such as an article by a politician, do not move the dot.
3. Two artificial intelligence models read every text, from two different companies. They say only which issue it touches, which side it takes, and which of the author's sentences shows it.
4. Where they disagree, or only one of them found a position, a third model decides, from a third company, without knowing which outlet it is.
5. We calculate with plain arithmetic. We count columnists, not texts, so that one prolific columnist does not become "the outlet".
6. The map is updated every night. The editors can withdraw an edition, and every withdrawal is recorded.

At no step does a model decide what is "left" or "right". The model says only what the author supports. Which side that counts towards is set by the list.

An example

A made-up outlet, "Outlet X", writes about the issue "The state should raise wages, through the minimum wage or collective agreements that apply to everyone, rather than leave them to the market". Support counts towards state intervention (−1) and opposition towards the free market (+1).

• The editorials: 3 texts, all in favour → −1. They count double, because they speak for the editors.
• Columnist A: 4 texts, 2 in favour, 1 against, 1 mixed → −0.25.
• Columnist B: 1 text against → +1.
• Columnist C: 1 text in favour → −1.

Together: [2 × (−1) + (−0.25) + 1 + (−1)] ÷ 5 = −0.45.

The same is done for every economic issue, and their average gives the outlet's place on the axis. For Outlet X the average is −0.31. On the map, positions are written three times larger, from −3 to +3, so Outlet X comes out at −0.9: clearly towards state intervention, but far from the extreme. Each outlet's page shows the texts that counted, with their sentence and a link to the original.

What the shade around an outlet means

We calculate each outlet's position 1,000 times, each time with a random sample of its columnists and their texts. On each axis separately, the shade spans the range where 9 in 10 of those positions fall.

• A small shade: plenty of consistent evidence.
• A large shade: little evidence, or columnists who disagree with each other.

On the map, the shade appears when you point at an outlet.

When we show no position

We would rather show no position than one that is not supported. The map then says:

• "Too few texts": the outlet took a position on fewer than half the axis's issues.
• "Too few columnists": fewer than three columnists took a position (the editorials count as one).
• "Depends on one columnist": without them, the position would fall outside its shade on one of the axes.
• "Publish news only": the outlet has no editorials or regular columnists. You can see it through its stance towards the government and its coverage.

We never put an outlet in the centre because we lack evidence. The centre means moderate or balanced positions, not "we don't know".

How we measure the stance towards the government

This is a different question from political positions, and it never goes on the map. An outlet can agree with the governing party on many things and criticise the government, or the opposite.

• We take the stories covered by at least 6 outlets.
• A model reads all the story's headlines together, without the outlets' names, and judges each one: positive, critical, neutral or "not about the government".
• Every headline judged positive or critical is read again by another model, from another company, which does not know the first judgement. The side counts only when the other model finds it too; otherwise the headline counts as neutral.
• A side does not count when a statement or act of a foreign government or an international body carries it, for example a threat from Ankara or praise from Washington or the Commission: it is not the outlet's judgement of the Greek government, and the headline counts as neutral. It counts as usual when a Greek speaker talks about a foreign matter, or when the headline reports the government's own act with a foreign party, such as a repayment or a meeting.
• For each outlet, only the stories in which its headlines took a side count. A story counts as positive when the outlet has more positive than critical headlines on it, critical the other way round; neutral headlines do not cancel a headline that takes a side.
• The grade, from −3 to +3, comes from the ratio of the two sides: ±1 when one side has at least twice as many stories as the other, ±2 four times, ±3 nine times, otherwise 0. At least 20 stories with a side are needed. Next to the grade, the page shows in how many of its stories the outlet took a side.

A "positive" headline means it puts forward a government success, not that the outlet agrees. Whenever the government changes, the count starts again.

What changed on 22 October 2026. In a check of 240 headlines with a side, 15% of the critical and 21% of the positive ones were plainly wrong: headlines about a foreign government, headlines in which the government speaks about the opposition instead of being judged, plain reports read as taking a side. By decision of the editorial team, every side is now read again by another model, and counts only when the two agree. In the same check, the plain errors fell to 3% of the critical and 7% of the positive headlines. About half of the stories with a side count now, so some smaller outlets have fewer than 20 for the moment and get no grade. Before it was published, the new measure was computed beside the old one from the same judgements: on 22 October 2026 it gave 10 pro-government, 31 neutral and 20 critical outlets, against 10, 42 and 17 under the previous rule; 19 outlets had fewer than 20 stories with a side, against 11.

And later the same day. By decision of the editorial team, the sides a foreign government or an international body carries no longer count. Of the 6,555 sides that counted since 1 September, 249 left: 185 positive ones, mostly praise from the United States, the Commission, the IMF and the World Health Organization, and 64 critical ones, mostly threats from Turkey. The measure then gave 9 pro-government, 29 neutral and 21 critical outlets, against 10, 30 and 21 without this rule; 21 outlets had fewer than 20 stories with a side, against 19.

What changed on 14 October 2026. Until then every relevant story counted: (positive − critical) ÷ all, with limits at 0.15, 0.40 and 0.70. About two in three stories about the government are plain reports, so an outlet's group also depended on how many such stories it publishes, not only on how it comments. By decision of the editorial team, only the stories in which the outlet's headlines take a side count now. Before it was published, the new measure was computed beside the old one from the same judgements.

A second model, from another company, checks the judgements on 300 headlines. The check is done by a model, not by people. For the measure published today it passed on 11 October 2026; while the check has not passed, the figures are provisional. The state of the check is shown at the top of this page.

When the editors correct the judgement on a headline, for example after an expert's “Wrong”, the correction counts and the headline shows its reason. The figures are the model's judgements, as the other model confirms them, together with these corrections; the check on 300 headlines measures the judging model alone.

The groups and the coverage bars

By their stance, outlets fall into three groups: critical, neutral and pro-government. While an outlet does not have enough measured stories, its group is the editors' estimate. Each story's bar shows how many outlets from each group covered it.

Blind spots

A blind spot is a story covered almost only by pro-government outlets, or almost only by critical ones. We count who covered it, among the outlets we track and are currently ingesting.

We mark it only if at least 3 outlets covered it, 24 hours have passed so the other side has had time to write, such an absence would happen by chance less than 1 time in 20, and we searched and found no text from the missing side. Finally an artificial intelligence model judges whether the story tells the reader something political; those that do not, such as a lottery draw, are not shown.

Who does what

Every model decision is stored with the reason and the sentence that supports it. Every month the third model re-judges a random sample of the positions on which the two readings agreed, to make sure the process is followed correctly.

Limits of our method

Every measurement rests on choices. We cannot claim these choices are 100% objective. We do not believe they can be. We are subjective beings by nature. But with a clear process for our choices, we can at least all see our biases.

• The issues we chose shape the axes. That is why the list is public.
• Models make mistakes, for example with irony.
• The map looks at the last 12 months, so a sudden change of line shows slowly.
• The map does not see which stories an outlet chooses. Coverage and blind spots show that.

Frequently asked questions

Do you use artificial intelligence?

Yes, to read many headlines and some texts. What each model does is in the table "Who does what".

Why not simply "Left, Centre, Right"?

Because one bar hides differences. A national-conservative outlet that asks for a lot of state in the economy would come out "left" on a bar where the economy counts for more. On the map it shows as it is: close to state intervention and close to the conservative end.

What does a position on the map not mean?

It is not a grade of quality or reliability. It does not say which outlet is right, or which party it supports.

Why are some outlets not on the map?

Some do not publish opinion of their own: they have no editorials or regular columnists. We tried to place them from the headlines and from the text of their news reports, but news reports rarely take a position. You can see these outlets through their stance towards the government and their coverage.

Others publish opinion, but we do not yet have enough of their texts or columnists to show a position. The map fills in slowly, as we read new texts.

Can an outlet's position change?

Yes. The map is updated every night from the texts of the last 12 months, so a position changes slowly, as new texts come in and old ones drop out.

Do readers' opinions count?

Readers' ratings appear next to ours once there are at least 30 for an outlet, but they never move the dot. In "Where do you stand?" you can also see your own place on the map. Only you can see your answers.

Transparency, corrections, right of reply

We publish the list of issues, the positions with the sentence that supports them and a link to the original, and the formulas of the calculation.

The detailed rules, the full list of issues, the formulas and the sources are in the technical methodology.

Found a mistake? Anyone can dispute a position with "Dispute" on each outlet's page. A person on the editorial team reviews the request, not a model. Every correction goes into the corrections log: what changed, when and why.