Case study
Loot Crawler
Read the chest before you open it.
Loot Crawler is a mobile game about expeditions into procedurally generated dungeons. You collect loot against the clock with a limited number of keys, and the real skill is knowing which chest is worth opening.
The challenge
Opening chests shouldn’t be a coin flip.
Without information, a loot game turns into random clicking. We wanted the opposite: no combat and no reflex tests, just one decision that carries the whole game. Which chest gets your key, when you can’t be sure what’s inside?
The idea
Investigation. Every chest hides three clues: weight, sound and magic. Studying a chest costs precious seconds, and cheaper gear gives noisier readings. The player weighs the evidence, then spends a key or saves it, like an appraiser rather than a gambler. Between runs: sell the loot, upgrade the gear, go deeper.
What we built
The Investigation system
Three hidden clues per chest, read with noise that depends on the player’s investigation level. The clues stay consistent: a chest investigated twice shows the same thing.
Time, keys and a backpack
Every run balances three limits: the clock, a small supply of keys and a backpack that fills up. Deciding when to head back is part of the game.
Reproducible dungeons
Dungeons are generated from a seed with deterministic randomness, so the same seed always builds the same dungeon. That makes every run testable and every balance change measurable.
A balancing bot in CI
A bot plays thousands of simulated expeditions with different strategies, and balance thresholds act as a gate in continuous integration.
The first Unity screens
A dungeon map, tap-to-move and a HUD that shows the cost of heading back, running on temporary procedural sprites.
How we built it
The rules of the game live in a pure C# core with no reference to the engine. The same source compiles twice, in Unity for the game and in .NET for tests and the balancing bot, and Unity refuses to build if anyone imports the engine into the core. That is why thousands of expeditions can be simulated from the console, without opening the editor.
When the simulation disagreed with the design document, the document lost: simulated runs on the actual game code corrected several of the original numbers.
- AI in the process
- Development runs on a team of coordinated AI agents, each owning one domain: design, programming, QA. Every task follows the same path (question, options, decision, draft, approval) with a person making the calls.
Stack
Status & what’s next
Loot Crawler is in production and has not been released. The core systems are built and covered by tests, and the Unity layer is coming together screen by screen. There is no release date yet.
What’s next
- Chest cards, investigation and the chest-opening moment on screen.
- Saving, the base, the shop and upgrades.
- Final key art, sound and playtests on real phones.
