Chimera
A graph of actors with profiles and the links between them, tied to the news. Its own graph engine.
Chimera keeps the actors: states, people, organisations, companies, groups, places, events, assets. Each one has a type, aliases, and relations to other actors, weighted by how often they appear together in the news. Where there is enough material, a local language model writes a profile: motivations, traits, leverage points.
The graph runs on an embedded engine chosen for the purpose and switched over on 31 May 2026. Actors arrive from Retina: every entity the news pipeline extracts becomes a candidate, and a spike around a name triggers a fresh profile. Ownership and sanctions registers sit alongside, several hundred thousand edges, but they barely touch the news actors: measured in July, 53 links could be drawn both ways. That join is parked, not built.
What is not yet there, stated plainly. A profile cannot be traced back to the news articles behind it; the pipeline does not keep that link, and a checking tool cannot either. Where a web search fed a profile, the page addresses are kept. An automatic merge once left about one actor in thirteen carrying another actor’s name as an alias; after two clean-ups in September, about one in 250 still does, and the path that caused it is closed. A layer of confirmed same-actor links holds 84 entries and has not grown since 1 September. Six large state actors have been curated by hand and are locked against automatic overwriting. The numbers below are read from the running system on the day of the snapshot; the graph grows while you read.
What goes in
Snapshot 8 October 2026Actors arrive from Retina: every entity the news pipeline extracts becomes a candidate. What the graph adds is structure: a type, aliases, relations to other actors weighted by how often they appear together, and, for 11,207 of the 17,713, a profile written by a local model.
Actors by type
- not yet typed5,000
- people3,873
- cities1,744
- organisations1,518
- companies1,407
- regions861
- assets659
- events620
- institutions505
- countries388
- concepts238
- groups117
- platforms111
- parties105
"Not yet typed" is honest: at least 5,000 actors are in the graph but have not been classified.
Relations
- relates997,098
- expressed5,006
- about5,006
- mentions2,217
Bars on a log scale. "Relates" is co-mention in the news; "expressed" and "about" carry stances; "mentions" links to articles.
What comes out
Most connected actors
| actor | type | relations |
|---|---|---|
| United States | country | 3,753 |
| Russia | country | 2,862 |
| Iran | country | 2,309 |
| Argentina | country | 1,966 |
| Ukraine | country | 1,747 |
| Israel | country | 1,549 |
| Israel | country | 1,287 |
| Hezbollah | group | 1,252 |
| Venezuela | country | 1,151 |
| Bloomberg | media | 982 |
| TikTok | platform | 904 |
| platform | 752 | |
| Hamas | group | 667 |
| Israel | country | 664 |
| Press TV | media | 619 |
| Claude | platform | 595 |
One actor and its neighbourhood: United States
The strongest relations of a single actor, weight from co-mention in the news.
- Islamic State0.87
- Iran0.30
- Mossad0.26
- Newark0.24
- Kennedy Center0.23
- MBDA0.20
- Gripen0.19
- Elbit Systems0.19
- Тунгуска-М10.18
- Группа войск «Центр»0.17
- Pakistan Tehreek-e-Insaf0.15
- US CENTCOM0.14
- Iranian Parliament0.14
- F-35 Lightning II0.13
- Strait of Hormuz0.12
- Patriot0.12
- Україна0.12
- MQ-20 Avenger0.11
One type at a time
The bar chart above is a headcount. This is what each type actually contains: the best connected actors of that type, with the number of relations behind each name, and for the leading one the relations themselves.
people
3,873 in the graph- Javier Rubio406
- Mark Rutte364
- Kyedae52
- Rima Hassan43
- Rex Tan31
Strongest relations of Javier Rubio
- JP Morgan · companies0.8238
- Daniel Jalkut · people0.7889
- Dario Amodei · people0.3031
- Sundar Pichai · people0.2954
- FIFA World Cup · events0.2015
- DeepMind · organisations0.1388
cities
1,744 in the graph- Hamburg389
- Palo Alto189
- Bagicz156
- Lampedusa153
- Uvira139
Strongest relations of Hamburg
- Wales · regions0.2018
- Magdeburger Weihnachtsmarkt · events0.1303
- Pilbara · regions0.1264
- Alice Springs · cities0.1227
- OHCHR · not yet typed0.1224
- Siegburg · cities0.1218
organisations
1,518 in the graph- International Rescue Committee140
- International Council on Ethics111
- All Girls Foundation for Development29
- Pharmaceuticals Action Committee0
- Humanitarian Needs Response Platform0
Strongest relations of International Rescue Committee
- UN Women · not yet typed0.2417
- OCHA · organisations0.1689
- UNFPA · institutions0.1633
- Protection Cluster · organisations0.1402
- Wale Osofisan · people0.125
- Administración Pública Nacional · not yet typed0.0985
companies
1,407 in the graph- Shell317
- AI21 Labs180
- Cisco Systems125
- Bain Capital124
- Quantum Systems GmbH94
Strongest relations of Shell
- AstraZeneca · companies0.48
- Ocado Group · companies0.3444
- Manchester City · cities0.3333
- JP Morgan · companies0.2845
- Lloyds · companies0.2656
- Nationwide · companies0.2488
regions
861 in the graph- Central Africa282
- Krasnodar Krai238
- Poltava Oblast190
- Maputo Province117
- Komodo National Park107
Strongest relations of Central Africa
- Cornwall · regions0.1521
- EFAS · companies0.1179
- International Paper Company · organisations0.105
- European Flood Awareness System · institutions0.0985
- El Niño · events0.0968
- Côte d'Ivoire · countries0.0886
assets
659 in the graph- Shahed174
- Iskander Muhammad0
- Trove Token0
- Pendle Token0
- AAVE Token0
Strongest relations of Shahed
- 2025 Belgorod incursion · events0.101
- Islamic Revolutionary Guard Corps · institutions0.0758
- Luhansk People's Republic · countries0.0612
- Saada Governorate · regions0.0561
- IRGC · not yet typed0.0515
- General Staff of the Armed Forces of Ukraine · organisations0.0362
events
620 in the graph- China Development Forum100
- Iran War68
- Karachi Fire0
- US-Israeli-Iran Conflict0
- Ukraine EU Accession0
Strongest relations of China Development Forum
- République centrafricaine · countries0.2121
- UNHCR · institutions0.1699
- IFRC · not yet typed0.16
- Fédération Internationale de Football Association · organisations0.1587
- WFP · institutions0.1533
- Health Cluster · organisations0.1517
institutions
505 in the graph- Senate Banking Committee196
- UK Parliament169
- Federal Office for the Protection of the Constitution158
- Executive Branch of the United States Government126
- Security Service of Ukraine104
Strongest relations of Senate Banking Committee
- Genius · organisations0.2091
- Securities and Exchange Commission · organisations0.2084
- CLARITY Act · concepts0.1984
- Commodity Futures Trading Commission · organisations0.1321
- Coinbase · companies0.1053
- Congress · institutions0.0974
countries
388 in the graph- United States3,753
- Russia2,862
- Iran2,309
- Argentina1,966
- Ukraine1,747
concepts
238 in the graph- International Monetary System130
- World Order24
- Transatlantic0
- Humanitarian Situation Monitoring0
- Pig Butchering0
groups
117 in the graph- Hezbollah1,252
- Hamas667
- Boko Haram366
- La Inmaculada193
- Palestinian Islamic Jihad100
platforms
111 in the graph- TikTok904
- WhatsApp752
- Claude595
- Kick381
- npm256
parties
105 in the graph- Green Party458
- CDU457
- FDP323
- CSU292
- Republican Party220
media outlets
101 in the graph- Bloomberg982
- Press TV619
- Agence France-Presse550
- Forbes397
- Axios361
movements
23 in the graph- Make America Great Again322
- Kirchnerismo224
- Southern Transitional Council108
- Wolpack0
- Climate Activism0
Two things this does not claim. The relation type is almost always RELATES, which means "appeared together in the same reporting", not a stated link between the two. And the quality varies by type: for an aid organisation the relations read like a sector map, for others they are looser. Shown as measured, not curated. Examples are drawn from actors that carry a profile, so a type with many entries can still show few.