How the board works
What the Global Situation Board is, who writes it, how often it changes, and how far to trust it.
What it is
The board tracks 56 flashpoints and 30 actors of concern around the world: wars, crises and standoffs, and the states, armed groups, criminal organisations and hacking and influence operations behind them. Every entry has a threat level from 1 to 5, a written assessment, and dated events with links to their sources. Flashpoints also carry a status, a 30-day trend, the indicators that would signal a change, and the dates coming up.
Who writes it
AI research agents built on Claude, Anthropic’s model. They search the web, read news reports, official statements and research organisations’ assessments, and write in a fixed format that a validation script checks before anything reaches the board. The script checks structure (dates, links, allowed values), not truth.
Nobody reviews an update before it is published. The board has an editor, who set up the process and does not approve individual updates. The editor can step in, by adding an entry (usually one the research agents proposed), by asking for a research review between the weekly ones, or by setting a level by hand.
How it updates
- Every morning, headlines from 90 news feeds and searches are pulled and matched to entries by keyword, and a short morning note picks out the overnight movers. The morning note is written from the headlines alone, without opening the articles, and never changes a level. Matching is automatic, so some headlines under an entry will be off-topic.
- Weekly, normally on Mondays (a review can run late or be rerun), a research review reassesses the entries, with one agent per region. Each agent rewrites the assessment, adds dated events, each with a link to its source, and may change the level. A review moves a level at most one step unless a war starts or ends, a ceasefire takes hold, a coup succeeds or a state collapses.
- New entries come from the research agents, who propose a flashpoint or actor when something important falls outside every existing entry. Proposals are added at the next weekly review, which writes the first assessment or drops the entry if it turns out not to be warranted. The editor can also add an entry directly; its first assessment then comes from a research review run between the weekly ones. Entries no longer worth tracking can be retired by a weekly review.
- Reviews run between the weekly ones are labelled “out-of-cycle review” in each entry’s level history and on the changes page, which lists every change of level, status or trend and every addition and removal.
The threat scale
The level answers “how severe is this over the next 30 days”, not “how much happened this week”. A bad week in a war already at 4 or 5 keeps its level.
| Level | Name | Meaning |
|---|---|---|
| 5 | Severe | large-scale war or imminent major escalation with global consequences |
| 4 | High | major active conflict, or serious risk of interstate escalation |
| 3 | Elevated | active low/mid-intensity conflict or crisis with real escalation risk |
| 2 | Guarded | political tension or low-level violence |
| 1 | Low | latent dispute, no violence expected near term |
A flashpoint’s status is one of: Active war, Escalating, Tense/Stable, De-escalating, Frozen, Watch. Its trend is the direction over the past 30 days: escalating, stable or easing. Actors carry a level only.
Sources and caveats
Every assessed event links to its source, and each entry lists the sources behind its assessment. The agents are told to mark events that rest on one outlet as “single-sourced”, figures outlets disagree on as “contested”, and claims read only from a search result, because the page could not be opened, as “snippet only”, and to hedge such claims in the written assessment (“reportedly”, “an outlet reports”). They do not always do so, and nothing checks the assessments for it. Some major outlets block automated reading, so research sometimes relies on snippets or on secondary reports.
When a review finds nothing new on an entry, it keeps the assessment and moves the assessed date, so the date shows when an entry was last reviewed, not when something last happened; the dated events show what actually changed.
The agents can be wrong. They can misread a source, miss a development, or be misled by an out-of-date page, and they can only use what is public. Each entry shows the date of its last assessment and each page the time of the last headline pull; if an update is missed, those dates show it.
Levels set by hand
The editor can set a level by hand between reviews. The board then marks that entry “Editor’s level”, and the next weekly review decides on its own evidence whether the level stands: the reviewing agent is told to give the editor’s number no weight. The changes page records whether the review upheld or overrode it. Levels set by hand are not part of an entry’s level history, which records research reviews only.
What it is not
This is not news reporting, intelligence, or advice for travel, safety, investment or any other decision. For travel, use your government’s official advisories. Actors are named by the government or organisation and, where it helps identify them, the leader associated with it; being listed reflects the agents’ reading of public reporting, not a legal finding about anyone.
Headlines link to their publishers; the board does not host articles.
Data and feeds
- Briefs feed (Atom): weekly reviews and morning notes.
- Changes feed (Atom): changes of level, status or trend, and entries added or removed.
- watchlist.json: every entry as it stands now.
- history.json: each entry’s level, status and trend after every research review.