How TravintSecurity works
Six steps from public sources to the risk score you see on screen.
The short version
Every country on the map has its risk assessed across nine categories. Structural facts come from authoritative public data — statistics published by the UN, the WHO, specialised conflict research programmes, press-freedom organisations, and similar bodies. Current events are continuously gathered from a curated set of news outlets. A deterministic scoring engine — not a language model — converts facts into the colour tiers you see on the country panel. Language models help us read, summarise, and explain what’s happening, but the scores themselves come from rules.
The six steps
Gather structural country data
Every country profile starts with anchor data from named authoritative sources — homicide rates, health-system effectiveness, press freedom, armed-conflict deaths, natural- hazard exposure, government response capacity, and a dozen similar indices. These are updated annually, or whenever the publisher releases new numbers.
If a source disagrees with our score for a category, the source wins. We don’t override structural data unless an acute event justifies it.
Listen to the news continuously
A monitoring pipeline pulls headlines from around eighty curated news outlets, from wire services and major broadcasters to country-specific papers and conflict specialists. Every hour, new articles are triaged, classified into a structured event taxonomy, and tagged with country, date, severity, and (where relevant) a specific region.
Cross-check before anything is believed
A single news report is a rumour. We require two independent sources — or one source from an established wire service or major outlet — before an event counts toward a score change. We track the source history of every event. Reports that fail the bar sit in a holding area, waiting for corroboration.
Score by rules, not by gut
A deterministic scoring engine takes all the facts — structural anchors plus confirmed events — and applies explicit thresholds to produce a colour tier per category. A homicide rate above thirty per hundred thousand forces Crime to at least RED. A certain number of fatal terrorist attacks in a twenty-four- month window moves Terrorism past a threshold. Fighting on a country’s own territory forces Armed Conflict upward. The rules are open-source and covered by hundreds of automated tests.
The engine can’t be talked out of a result. If the data says PURPLE, the score is PURPLE.
Apply identity lenses on top
Country scores describe risk to a generic international traveller. Identity lenses — for Jewish / Israeli travellers, solo women, and more to come — layer additional risk on top. Lenses can only raise a category above base, never lower it. Each lens has its own scoring rubric built from specific structural data: documented antisemitic-incident patterns, gender-based violence statistics, enforced legal codes, and so on.
Explain in plain language
Once the scores are calculated, a language model writes the narrative — the “why” behind the score, the key watch factors, the specific advice — grounded in the same evidence the engine used. When a score changes because of a specific event, the event itself is cited in the explanation and available as a drill-down.
How we handle mistakes
Language models hallucinate. Our pipeline misclassifies things. Our regional breakdowns sometimes get geography wrong. We have a defence in depth:
- Source-trust filters.News that only comes through automated aggregators without a human-edited source doesn’t count toward scoring.
- Framework constraints. Some scores have hard ceilings that cannot be exceeded. Some categories have explicit scope rules — for example, organised-crime violence is always counted as Crime, never as Armed Conflict.
- Geographic discipline.Events tagged to one region can’t raise scores in another region without explicit justification.
- Traveller flagging. Every score page has a flag button. Flags are pre-screened automatically, then reviewed by a human. Accepted flags patch the score directly.
- Audit trail.Every score change is logged with its driving evidence. We can always answer “why did this score change”.
Tiers — what you see
Each category and the country total show one of five colours:
- Low — no meaningful risk above baseline.
- Elevated — some risk factors present; aware but not limiting.
- Significant — meaningful risk; precautions required.
- High — serious documented risk; plan carefully.
- Extreme — catastrophic risk; travel strongly inadvisable.
What you get (free vs paid)
Free tier: the total country score, a short intelligence summary, a note on whether regional differences apply, and safety-critical natural-hazard alerts for any country with an active event.
Paid tier: the full per-category breakdown with explanations, personalised travel advice, historical score trends with sparklines, admin-level regional detail, event drill-down per region, identity-lens layers, and a way to flag inaccuracies directly into our review queue.