How these scores are computed

Scores are computed, not chosen. No person selects a country's rating, and no language model selects one either. The full specification and the engine itself are open source, so any published number can be recomputed from the published data.

Seven dimensions

Each is scored 0 to 100 independently, where higher means more risk.

Border device risk: How likely is my device to be searched, imaged, or seized at the border, and can I be compelled to unlock it?
Surveillance and interception: How comprehensively are communications intercepted, retained, and monitored, and is commercial spyware in use?
Encryption and VPN legality: Are VPNs and strong encryption lawful, and can I be required to install an application or hand over keys?
Network threat: How hostile is the network environment to a foreign visitor, from state actors and criminals alike?
Speech and legal exposure: What legal risk attaches to what I have said online, or say while I am there?
Data jurisdiction: What happens to corporate or client data I carry into the jurisdiction?
Connectivity continuity: How likely am I to lose access to the internet, or to the specific tools I depend on?

They are deliberately not summed into one number on the country page. The traveller worried about legal exposure for what they have posted is often not the one carrying client data.

How a dimension score is built

A slow moving baseline from legislation, standing advisories and published indices, plus an overlay of recent incidents that fades as they age. How fast depends on what the incident describes. An outage halves in a month, because connectivity recovers. A border search or a prosecution halves in a year, because a state that did it once tends to do it again. The overlay is capped either way, because volume of news coverage is not the same as risk.

Evidence goes stale

Old evidence loses weight rather than counting forever. Primary legislation fades over about five years, enforcement practice within one, and anything over ten years old is dropped entirely and reported as excluded. Where a law is repealed or a block lifted, the signal stops counting immediately rather than fading.

Confidence, and what it is for

Coverage is not uniform. Every score carries a confidence value combining how many independent sources contribute, how recent they are, and whether they agree. A score built on two ageing sources should not present itself like one built on nine current and agreeing ones, so it does not.

Countries below the threshold are marked low confidence on their page. They are published rather than hidden, because a weak signal openly marked weak is more useful than a gap.

Your passport changes the answer

Government travel advice is written for a specific citizenry. UK advice is written for UK nationals and US advice for US nationals, and the two routinely disagree about the same country. Selecting your origin on a country page weights advice written for you more heavily, without discarding the rest.

Known limitations

Seven dimensions is a simplification of a domain that does not naturally decompose into seven things.
Normalising heterogeneous sources onto one scale involves judgement. Those choices are ours, they are published, and they are arguable.
English language sources dominate, which biases observation toward countries anglophone media cover.
Absence of evidence is treated as lower confidence, never as lower risk. A country with no reported device searches may have none, or may have nobody free to report them.
No primary research. Everything derives from published sources and inherits their errors.

Model version 0.12.0.