On July 8, 2026, we took 1,012 world-news headlines about 11 countries and asked each one a single question: who is acting here? A named person, a named institution — or the whole country, as one faceless giant?
| Action type | Country-monolith share | |
|---|---|---|
| Violence | 89% · 54/61 | |
| Policy, sanctions | 81% · 29/36 | |
| Diplomacy | 73% · 11/15 | |
| Speech | 46% · 16/35 |
We are not looking for culprits, and we pass no verdicts. We return names to their place: every decision has an author, and precise language names that author. Everyone has the right to an honest position — honest with the facts, under the author’s name. “Putin ordered” can be verified against a document. “Russia wants” is unverifiable in principle — the claim has no subject. That is why the monolith is so convenient for lies and costs the world so dearly.
This has been shown experimentally: when an action is attributed to a country as a whole, hostility toward that country’s people rises; when it is attributed to a specific actor — it falls (Wang, Li, Huang & Hsee, Judgment and Decision Making, 2025).
This is a pilot. One day (July 8, 2026), one aggregator (Google News), English-language headlines, 11 countries, a symmetric sample.
How we counted. Two-stage classification: rules resolve clear-cut cases; every ambiguous headline was reviewed individually by an LLM. The scale: individual / institution / capital-as-metonymy / country-as-monolith. Full dataset — 1,012 rows, each with its class and verification stage: agentivity_pilot_2026-07-08.csv.
What is still missing. Independent validation of the annotation — it will be the first task for the project’s collaborators. Numbers from a single day are a direction, not a law: verification across a longer series lies ahead.
That is the language of peace. Not a new language — just a precise one.