Presets
Playback
Trip
Transport
Connections
Preparation
Layers
Model check
The science
1. Distance on a sphere
Each leg length is the great-circle (haversine) distance d = 2R asin sqrt( sin2(dphi/2) + cos phi1 cos phi2 sin2(dlambda/2) ) with R = 6371 km. The model multiplies it by a mode-specific route factor (flight 1.05, train 1.20, coach and car 1.25) because real routes are not geodesics. The map is a flat equirectangular picture (x scaled by cos 48.5 degrees), so arcs drawn on it are only schematic; the numbers come from the sphere, not from the picture.
2. Choosing the order of the stops (travelling salesman)
For n cities there are (n-1)! round trips from home. Nearest-neighbour builds one greedily; 2-opt then removes any pair of legs that cross, replacing (a,b)+(c,d) by (a,c)+(b,d) whenever d(a,c)+d(b,d) < d(a,b)+d(c,d), until no swap helps. This finds a local optimum, not always the global one. For up to 9 cities the page also tries every order (brute force, Heap's algorithm) and, independently, solves the exact problem with the Held-Karp dynamic programme (O(n2 2^n)); both must agree. The "gap" is (2-opt - optimum) / optimum. You can minimise distance, time or CO2: with mixed modes the best order can change.
3. CO2 per passenger-km
CO2 = sum over legs of (great-circle km x route factor x emission factor). Typical emission factors used here (g CO2 per passenger-km): flight 150, train 35, coach 30, car 170 per vehicle-km divided by the number of people in the car. These are round, illustrative values of the order found in public tables (for example UK government conversion factors); the real figure depends on the year, aircraft, load factor, grid mix and fuel. The optional aviation multiplier (x1.9) stands for non-CO2 effects (contrails, NOx) which are scientifically uncertain (roughly 1.7 to 3). The 2.3 t per person per year reference is a commonly cited illustrative personal budget, not a legal limit.
4. Missed connections
The arrival delay of each leg is modelled as a normal variable D ~ N(mu, sigma2) (flight 8 +/- 22 min, train 3 +/- 12, coach 5 +/- 15, car 4 +/- 18; illustrative). A transfer with scheduled buffer B fails when D > B, probability Q((B - mu)/sigma) where Q(z) = 0.5 erfc(z/sqrt 2). With independent transfers, P(at least one miss) = 1 - prod(1 - p_k). A Monte Carlo run (seeded random numbers, Box-Muller normals) draws whole trips and counts them; it must match the formula within about 4 standard errors sqrt(p(1-p)/N). Simplifications: real delays are skewed (a long right tail), correlated (one storm hits several legs) and cancellations are ignored; the normal model also allows negative delays (early arrival). The buffer is the same at every stop.
5. Readiness gate
The preparation checklist is a weighted score (illustrative weights). If it is below the chosen threshold the traveller stays at home. It is a planning aid, not legal or medical advice: always check official entry and health rules for your passport and destination.
What is not modelled
Ferries and tunnels, schedules and timetables, prices, border queues, seasonal winds, aircraft type and distance-dependent fuel burn, rail electricity mix per country, and the human value of time. Times assume a flight overhead of 2.5 h (airport and transfer), train 0.7 h, coach 0.4 h, car 0.2 h. Land outlines: Natural Earth 110 m (public domain), simplified. Made with Blender (map table, vehicles, baked textures) and Three.js.