North Is Not Up

An orienteering game with no blue dot

North Is Not Up is a single-player 3D navigation game about working out where you are instead of following an icon that already knows.

The paper map never receives the runner’s position. There is no minimap, trail, route arrow, objective beacon, distance readout or permanent digital heading. You read contours and symbols, compare them with the landscape, choose a route and decide when the ground is telling you that you have gone wrong.

The compass works as a physical baseplate compass rather than a shortcut. On the map, you place it over the leg you want to travel, aim the direction-of-travel arrow and turn the bezel until its orienting lines match map north. Close the map, raise the compass and turn until red is in the shed. Any error in the setup comes with you.

That gap between a flat map and a place seen from ground level is the whole game.

Five different navigation challenges

• Woodland Training Area teaches map orientation, bearings and pace counting in a compact 100 × 100 metre landscape.
• Hawthorne Ridge is a timed six-control course with electronic punches, splits and finish results.
• Brackenford Crossing expands to eight controls across a 640-metre landscape where choosing the right river crossing matters.
• Black Crag Rescue replaces the standard course with a 900 × 900 metre mountain search. You follow clues, listen for whistles, stabilise an injured walker and report their grid square before the weather closes in.
• Ravenwood Score Challenge gives you 22 minutes to choose your own route through a 1.2-kilometre estate and collect as many of its 12 controls as possible.

The courses share the same navigation rules, but ask for different kinds of judgement: accurate bearings, route choice, relocation, search planning and deciding when to turn back.

The map is part of the world

Each course is an authored landscape with its own relief, paths, water, woodland, buildings, ruins and mapped features. The topographic maps are generated from the same terrain data used by the playable world, so a contour, stream bend or bridge is meant to survive the journey from paper to 3D.

Controls stay small and ordinary. They do not glow through trees or announce themselves from a distance. Wildlife moves through the courses, footsteps change with the surface underfoot, and later landscapes use fog, water and dense vegetation to alter what can be seen and heard.

Progression, a manual save slot, course timing, control splits and results all exist around the navigation without replacing it.

How I’m building it

The game is built in Godot with GDScript. Blender is used for assets such as the layered compass, while custom authoring tools bake fixed terrain, map imagery, collision, vegetation and landmarks from shared source data.

As the landscapes have grown from a 100-metre tutorial to kilometre-scale environments, the technical work has shifted towards keeping large worlds responsive and internally consistent. Ravenwood and Loch Ailsa stream detailed regions around the player while keeping course state, saves, map coordinates and distant scenery stable.

I’m using AI coding tools throughout the project, but the useful test is always physical: does the map match the ground, can a route be reasoned about, and does getting briefly lost feel fair?

Current state

There are five playable desktop courses with shared settings and campaign progression; Levels 1–4 also use one manual save slot. I’m currently building Loch Ailsa, a sixth landscape organised around islands, shorelines and open-water crossings. Its swimming, current, stamina and world-streaming systems are working, but it is still an exploration slice rather than a finished course.

Screenshots